Spaces:
Running
on
Zero
Running
on
Zero
zejunyang
commited on
Commit
•
2e4e201
1
Parent(s):
558ddd8
init
Browse files- src/audio2vid.py +255 -0
- src/audio_models/mish.py +51 -0
- src/audio_models/model.py +71 -0
- src/audio_models/torch_utils.py +25 -0
- src/audio_models/wav2vec2.py +125 -0
- src/models/attention.py +840 -0
- src/models/motion_module.py +388 -0
- src/models/mutual_self_attention.py +363 -0
- src/models/pose_guider.py +329 -0
- src/models/resnet.py +252 -0
- src/models/transformer_2d.py +396 -0
- src/models/transformer_3d.py +169 -0
- src/models/unet_2d_blocks.py +1074 -0
- src/models/unet_2d_condition.py +1308 -0
- src/models/unet_3d.py +673 -0
- src/models/unet_3d_blocks.py +861 -0
- src/pipelines/context.py +76 -0
- src/pipelines/pipeline_pose2vid_long.py +584 -0
- src/pipelines/utils.py +29 -0
- src/utils/audio_util.py +30 -0
- src/utils/draw_util.py +149 -0
- src/utils/face_landmark.py +3305 -0
- src/utils/mp_utils.py +95 -0
- src/utils/pose_util.py +78 -0
- src/utils/util.py +128 -0
- src/vid2vid.py +233 -0
src/audio2vid.py
ADDED
@@ -0,0 +1,255 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import ffmpeg
|
3 |
+
from datetime import datetime
|
4 |
+
from pathlib import Path
|
5 |
+
import numpy as np
|
6 |
+
import cv2
|
7 |
+
import torch
|
8 |
+
# import spaces
|
9 |
+
from scipy.spatial.transform import Rotation as R
|
10 |
+
from scipy.interpolate import interp1d
|
11 |
+
|
12 |
+
from diffusers import AutoencoderKL, DDIMScheduler
|
13 |
+
from einops import repeat
|
14 |
+
from omegaconf import OmegaConf
|
15 |
+
from PIL import Image
|
16 |
+
from torchvision import transforms
|
17 |
+
from transformers import CLIPVisionModelWithProjection
|
18 |
+
|
19 |
+
|
20 |
+
from src.models.pose_guider import PoseGuider
|
21 |
+
from src.models.unet_2d_condition import UNet2DConditionModel
|
22 |
+
from src.models.unet_3d import UNet3DConditionModel
|
23 |
+
from src.pipelines.pipeline_pose2vid_long import Pose2VideoPipeline
|
24 |
+
from src.utils.util import save_videos_grid
|
25 |
+
|
26 |
+
from src.audio_models.model import Audio2MeshModel
|
27 |
+
from src.utils.audio_util import prepare_audio_feature
|
28 |
+
from src.utils.mp_utils import LMKExtractor
|
29 |
+
from src.utils.draw_util import FaceMeshVisualizer
|
30 |
+
from src.utils.pose_util import project_points
|
31 |
+
|
32 |
+
|
33 |
+
def matrix_to_euler_and_translation(matrix):
|
34 |
+
rotation_matrix = matrix[:3, :3]
|
35 |
+
translation_vector = matrix[:3, 3]
|
36 |
+
rotation = R.from_matrix(rotation_matrix)
|
37 |
+
euler_angles = rotation.as_euler('xyz', degrees=True)
|
38 |
+
return euler_angles, translation_vector
|
39 |
+
|
40 |
+
|
41 |
+
def smooth_pose_seq(pose_seq, window_size=5):
|
42 |
+
smoothed_pose_seq = np.zeros_like(pose_seq)
|
43 |
+
|
44 |
+
for i in range(len(pose_seq)):
|
45 |
+
start = max(0, i - window_size // 2)
|
46 |
+
end = min(len(pose_seq), i + window_size // 2 + 1)
|
47 |
+
smoothed_pose_seq[i] = np.mean(pose_seq[start:end], axis=0)
|
48 |
+
|
49 |
+
return smoothed_pose_seq
|
50 |
+
|
51 |
+
def get_headpose_temp(input_video):
|
52 |
+
lmk_extractor = LMKExtractor()
|
53 |
+
cap = cv2.VideoCapture(input_video)
|
54 |
+
|
55 |
+
total_frames = int(cap.get(cv2.CAP_PROP_FRAME_COUNT))
|
56 |
+
fps = cap.get(cv2.CAP_PROP_FPS)
|
57 |
+
|
58 |
+
trans_mat_list = []
|
59 |
+
while cap.isOpened():
|
60 |
+
ret, frame = cap.read()
|
61 |
+
if not ret:
|
62 |
+
break
|
63 |
+
|
64 |
+
result = lmk_extractor(frame)
|
65 |
+
trans_mat_list.append(result['trans_mat'].astype(np.float32))
|
66 |
+
cap.release()
|
67 |
+
|
68 |
+
trans_mat_arr = np.array(trans_mat_list)
|
69 |
+
|
70 |
+
# compute delta pose
|
71 |
+
trans_mat_inv_frame_0 = np.linalg.inv(trans_mat_arr[0])
|
72 |
+
pose_arr = np.zeros([trans_mat_arr.shape[0], 6])
|
73 |
+
|
74 |
+
for i in range(pose_arr.shape[0]):
|
75 |
+
pose_mat = trans_mat_inv_frame_0 @ trans_mat_arr[i]
|
76 |
+
euler_angles, translation_vector = matrix_to_euler_and_translation(pose_mat)
|
77 |
+
pose_arr[i, :3] = euler_angles
|
78 |
+
pose_arr[i, 3:6] = translation_vector
|
79 |
+
|
80 |
+
# interpolate to 30 fps
|
81 |
+
new_fps = 30
|
82 |
+
old_time = np.linspace(0, total_frames / fps, total_frames)
|
83 |
+
new_time = np.linspace(0, total_frames / fps, int(total_frames * new_fps / fps))
|
84 |
+
|
85 |
+
pose_arr_interp = np.zeros((len(new_time), 6))
|
86 |
+
for i in range(6):
|
87 |
+
interp_func = interp1d(old_time, pose_arr[:, i])
|
88 |
+
pose_arr_interp[:, i] = interp_func(new_time)
|
89 |
+
|
90 |
+
pose_arr_smooth = smooth_pose_seq(pose_arr_interp)
|
91 |
+
|
92 |
+
return pose_arr_smooth
|
93 |
+
|
94 |
+
# @spaces.GPU
|
95 |
+
def audio2video(input_audio, ref_img, headpose_video=None, size=512, steps=25, length=150, seed=42):
|
96 |
+
fps = 30
|
97 |
+
cfg = 3.5
|
98 |
+
|
99 |
+
config = OmegaConf.load('./configs/prompts/animation_audio.yaml')
|
100 |
+
|
101 |
+
if config.weight_dtype == "fp16":
|
102 |
+
weight_dtype = torch.float16
|
103 |
+
else:
|
104 |
+
weight_dtype = torch.float32
|
105 |
+
|
106 |
+
audio_infer_config = OmegaConf.load(config.audio_inference_config)
|
107 |
+
# prepare model
|
108 |
+
a2m_model = Audio2MeshModel(audio_infer_config['a2m_model'])
|
109 |
+
a2m_model.load_state_dict(torch.load(audio_infer_config['pretrained_model']['a2m_ckpt']), strict=False)
|
110 |
+
a2m_model.cuda().eval()
|
111 |
+
|
112 |
+
vae = AutoencoderKL.from_pretrained(
|
113 |
+
config.pretrained_vae_path,
|
114 |
+
).to("cuda", dtype=weight_dtype)
|
115 |
+
|
116 |
+
reference_unet = UNet2DConditionModel.from_pretrained(
|
117 |
+
config.pretrained_base_model_path,
|
118 |
+
subfolder="unet",
|
119 |
+
).to(dtype=weight_dtype, device="cuda")
|
120 |
+
|
121 |
+
inference_config_path = config.inference_config
|
122 |
+
infer_config = OmegaConf.load(inference_config_path)
|
123 |
+
denoising_unet = UNet3DConditionModel.from_pretrained_2d(
|
124 |
+
config.pretrained_base_model_path,
|
125 |
+
config.motion_module_path,
|
126 |
+
subfolder="unet",
|
127 |
+
unet_additional_kwargs=infer_config.unet_additional_kwargs,
|
128 |
+
).to(dtype=weight_dtype, device="cuda")
|
129 |
+
|
130 |
+
|
131 |
+
pose_guider = PoseGuider(noise_latent_channels=320, use_ca=True).to(device="cuda", dtype=weight_dtype) # not use cross attention
|
132 |
+
|
133 |
+
image_enc = CLIPVisionModelWithProjection.from_pretrained(
|
134 |
+
config.image_encoder_path
|
135 |
+
).to(dtype=weight_dtype, device="cuda")
|
136 |
+
|
137 |
+
sched_kwargs = OmegaConf.to_container(infer_config.noise_scheduler_kwargs)
|
138 |
+
scheduler = DDIMScheduler(**sched_kwargs)
|
139 |
+
|
140 |
+
generator = torch.manual_seed(seed)
|
141 |
+
|
142 |
+
width, height = size, size
|
143 |
+
|
144 |
+
# load pretrained weights
|
145 |
+
denoising_unet.load_state_dict(
|
146 |
+
torch.load(config.denoising_unet_path, map_location="cpu"),
|
147 |
+
strict=False,
|
148 |
+
)
|
149 |
+
reference_unet.load_state_dict(
|
150 |
+
torch.load(config.reference_unet_path, map_location="cpu"),
|
151 |
+
)
|
152 |
+
pose_guider.load_state_dict(
|
153 |
+
torch.load(config.pose_guider_path, map_location="cpu"),
|
154 |
+
)
|
155 |
+
|
156 |
+
pipe = Pose2VideoPipeline(
|
157 |
+
vae=vae,
|
158 |
+
image_encoder=image_enc,
|
159 |
+
reference_unet=reference_unet,
|
160 |
+
denoising_unet=denoising_unet,
|
161 |
+
pose_guider=pose_guider,
|
162 |
+
scheduler=scheduler,
|
163 |
+
)
|
164 |
+
pipe = pipe.to("cuda", dtype=weight_dtype)
|
165 |
+
|
166 |
+
date_str = datetime.now().strftime("%Y%m%d")
|
167 |
+
time_str = datetime.now().strftime("%H%M")
|
168 |
+
save_dir_name = f"{time_str}--seed_{seed}-{size}x{size}"
|
169 |
+
|
170 |
+
save_dir = Path(f"output/{date_str}/{save_dir_name}")
|
171 |
+
save_dir.mkdir(exist_ok=True, parents=True)
|
172 |
+
|
173 |
+
lmk_extractor = LMKExtractor()
|
174 |
+
vis = FaceMeshVisualizer(forehead_edge=False)
|
175 |
+
|
176 |
+
ref_image_np = cv2.cvtColor(ref_img, cv2.COLOR_RGB2BGR)
|
177 |
+
# TODO: 人脸检测+裁剪
|
178 |
+
ref_image_np = cv2.resize(ref_image_np, (size, size))
|
179 |
+
ref_image_pil = Image.fromarray(cv2.cvtColor(ref_image_np, cv2.COLOR_BGR2RGB))
|
180 |
+
|
181 |
+
face_result = lmk_extractor(ref_image_np)
|
182 |
+
if face_result is None:
|
183 |
+
return None
|
184 |
+
|
185 |
+
lmks = face_result['lmks'].astype(np.float32)
|
186 |
+
ref_pose = vis.draw_landmarks((ref_image_np.shape[1], ref_image_np.shape[0]), lmks, normed=True)
|
187 |
+
|
188 |
+
sample = prepare_audio_feature(input_audio, wav2vec_model_path=audio_infer_config['a2m_model']['model_path'])
|
189 |
+
sample['audio_feature'] = torch.from_numpy(sample['audio_feature']).float().cuda()
|
190 |
+
sample['audio_feature'] = sample['audio_feature'].unsqueeze(0)
|
191 |
+
|
192 |
+
# inference
|
193 |
+
pred = a2m_model.infer(sample['audio_feature'], sample['seq_len'])
|
194 |
+
pred = pred.squeeze().detach().cpu().numpy()
|
195 |
+
pred = pred.reshape(pred.shape[0], -1, 3)
|
196 |
+
pred = pred + face_result['lmks3d']
|
197 |
+
|
198 |
+
if headpose_video is not None:
|
199 |
+
pose_seq = get_headpose_temp(headpose_video)
|
200 |
+
else:
|
201 |
+
pose_seq = np.load(config['pose_temp'])
|
202 |
+
mirrored_pose_seq = np.concatenate((pose_seq, pose_seq[-2:0:-1]), axis=0)
|
203 |
+
cycled_pose_seq = np.tile(mirrored_pose_seq, (sample['seq_len'] // len(mirrored_pose_seq) + 1, 1))[:sample['seq_len']]
|
204 |
+
|
205 |
+
# project 3D mesh to 2D landmark
|
206 |
+
projected_vertices = project_points(pred, face_result['trans_mat'], cycled_pose_seq, [height, width])
|
207 |
+
|
208 |
+
pose_images = []
|
209 |
+
for i, verts in enumerate(projected_vertices):
|
210 |
+
lmk_img = vis.draw_landmarks((width, height), verts, normed=False)
|
211 |
+
pose_images.append(lmk_img)
|
212 |
+
|
213 |
+
pose_list = []
|
214 |
+
pose_tensor_list = []
|
215 |
+
|
216 |
+
pose_transform = transforms.Compose(
|
217 |
+
[transforms.Resize((height, width)), transforms.ToTensor()]
|
218 |
+
)
|
219 |
+
args_L = len(pose_images) if length==0 or length > len(pose_images) else length
|
220 |
+
for pose_image_np in pose_images[: args_L]:
|
221 |
+
pose_image_pil = Image.fromarray(cv2.cvtColor(pose_image_np, cv2.COLOR_BGR2RGB))
|
222 |
+
pose_tensor_list.append(pose_transform(pose_image_pil))
|
223 |
+
pose_image_np = cv2.resize(pose_image_np, (width, height))
|
224 |
+
pose_list.append(pose_image_np)
|
225 |
+
|
226 |
+
pose_list = np.array(pose_list)
|
227 |
+
|
228 |
+
video_length = len(pose_tensor_list)
|
229 |
+
|
230 |
+
video = pipe(
|
231 |
+
ref_image_pil,
|
232 |
+
pose_list,
|
233 |
+
ref_pose,
|
234 |
+
width,
|
235 |
+
height,
|
236 |
+
video_length,
|
237 |
+
steps,
|
238 |
+
cfg,
|
239 |
+
generator=generator,
|
240 |
+
).videos
|
241 |
+
|
242 |
+
save_path = f"{save_dir}/{size}x{size}_{time_str}_noaudio.mp4"
|
243 |
+
save_videos_grid(
|
244 |
+
video,
|
245 |
+
save_path,
|
246 |
+
n_rows=1,
|
247 |
+
fps=fps,
|
248 |
+
)
|
249 |
+
|
250 |
+
stream = ffmpeg.input(save_path)
|
251 |
+
audio = ffmpeg.input(input_audio)
|
252 |
+
ffmpeg.output(stream.video, audio.audio, save_path.replace('_noaudio.mp4', '.mp4'), vcodec='copy', acodec='aac').run()
|
253 |
+
os.remove(save_path)
|
254 |
+
|
255 |
+
return save_path.replace('_noaudio.mp4', '.mp4')
|
src/audio_models/mish.py
ADDED
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Applies the mish function element-wise:
|
3 |
+
mish(x) = x * tanh(softplus(x)) = x * tanh(ln(1 + exp(x)))
|
4 |
+
"""
|
5 |
+
|
6 |
+
# import pytorch
|
7 |
+
import torch
|
8 |
+
import torch.nn.functional as F
|
9 |
+
from torch import nn
|
10 |
+
|
11 |
+
@torch.jit.script
|
12 |
+
def mish(input):
|
13 |
+
"""
|
14 |
+
Applies the mish function element-wise:
|
15 |
+
mish(x) = x * tanh(softplus(x)) = x * tanh(ln(1 + exp(x)))
|
16 |
+
See additional documentation for mish class.
|
17 |
+
"""
|
18 |
+
return input * torch.tanh(F.softplus(input))
|
19 |
+
|
20 |
+
class Mish(nn.Module):
|
21 |
+
"""
|
22 |
+
Applies the mish function element-wise:
|
23 |
+
mish(x) = x * tanh(softplus(x)) = x * tanh(ln(1 + exp(x)))
|
24 |
+
|
25 |
+
Shape:
|
26 |
+
- Input: (N, *) where * means, any number of additional
|
27 |
+
dimensions
|
28 |
+
- Output: (N, *), same shape as the input
|
29 |
+
|
30 |
+
Examples:
|
31 |
+
>>> m = Mish()
|
32 |
+
>>> input = torch.randn(2)
|
33 |
+
>>> output = m(input)
|
34 |
+
|
35 |
+
Reference: https://pytorch.org/docs/stable/generated/torch.nn.Mish.html
|
36 |
+
"""
|
37 |
+
|
38 |
+
def __init__(self):
|
39 |
+
"""
|
40 |
+
Init method.
|
41 |
+
"""
|
42 |
+
super().__init__()
|
43 |
+
|
44 |
+
def forward(self, input):
|
45 |
+
"""
|
46 |
+
Forward pass of the function.
|
47 |
+
"""
|
48 |
+
if torch.__version__ >= "1.9":
|
49 |
+
return F.mish(input)
|
50 |
+
else:
|
51 |
+
return mish(input)
|
src/audio_models/model.py
ADDED
@@ -0,0 +1,71 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import torch
|
3 |
+
import torch.nn as nn
|
4 |
+
import torch.nn.functional as F
|
5 |
+
from transformers import Wav2Vec2Config
|
6 |
+
|
7 |
+
from .torch_utils import get_mask_from_lengths
|
8 |
+
from .wav2vec2 import Wav2Vec2Model
|
9 |
+
|
10 |
+
|
11 |
+
class Audio2MeshModel(nn.Module):
|
12 |
+
def __init__(
|
13 |
+
self,
|
14 |
+
config
|
15 |
+
):
|
16 |
+
super().__init__()
|
17 |
+
out_dim = config['out_dim']
|
18 |
+
latent_dim = config['latent_dim']
|
19 |
+
model_path = config['model_path']
|
20 |
+
only_last_fetures = config['only_last_fetures']
|
21 |
+
from_pretrained = config['from_pretrained']
|
22 |
+
|
23 |
+
self._only_last_features = only_last_fetures
|
24 |
+
|
25 |
+
self.audio_encoder_config = Wav2Vec2Config.from_pretrained(model_path, local_files_only=True)
|
26 |
+
if from_pretrained:
|
27 |
+
self.audio_encoder = Wav2Vec2Model.from_pretrained(model_path, local_files_only=True)
|
28 |
+
else:
|
29 |
+
self.audio_encoder = Wav2Vec2Model(self.audio_encoder_config)
|
30 |
+
self.audio_encoder.feature_extractor._freeze_parameters()
|
31 |
+
|
32 |
+
hidden_size = self.audio_encoder_config.hidden_size
|
33 |
+
|
34 |
+
self.in_fn = nn.Linear(hidden_size, latent_dim)
|
35 |
+
|
36 |
+
self.out_fn = nn.Linear(latent_dim, out_dim)
|
37 |
+
nn.init.constant_(self.out_fn.weight, 0)
|
38 |
+
nn.init.constant_(self.out_fn.bias, 0)
|
39 |
+
|
40 |
+
def forward(self, audio, label, audio_len=None):
|
41 |
+
attention_mask = ~get_mask_from_lengths(audio_len) if audio_len else None
|
42 |
+
|
43 |
+
seq_len = label.shape[1]
|
44 |
+
|
45 |
+
embeddings = self.audio_encoder(audio, seq_len=seq_len, output_hidden_states=True,
|
46 |
+
attention_mask=attention_mask)
|
47 |
+
|
48 |
+
if self._only_last_features:
|
49 |
+
hidden_states = embeddings.last_hidden_state
|
50 |
+
else:
|
51 |
+
hidden_states = sum(embeddings.hidden_states) / len(embeddings.hidden_states)
|
52 |
+
|
53 |
+
layer_in = self.in_fn(hidden_states)
|
54 |
+
out = self.out_fn(layer_in)
|
55 |
+
|
56 |
+
return out, None
|
57 |
+
|
58 |
+
def infer(self, input_value, seq_len):
|
59 |
+
embeddings = self.audio_encoder(input_value, seq_len=seq_len, output_hidden_states=True)
|
60 |
+
|
61 |
+
if self._only_last_features:
|
62 |
+
hidden_states = embeddings.last_hidden_state
|
63 |
+
else:
|
64 |
+
hidden_states = sum(embeddings.hidden_states) / len(embeddings.hidden_states)
|
65 |
+
|
66 |
+
layer_in = self.in_fn(hidden_states)
|
67 |
+
out = self.out_fn(layer_in)
|
68 |
+
|
69 |
+
return out
|
70 |
+
|
71 |
+
|
src/audio_models/torch_utils.py
ADDED
@@ -0,0 +1,25 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
import torch.nn.functional as F
|
3 |
+
|
4 |
+
|
5 |
+
def get_mask_from_lengths(lengths, max_len=None):
|
6 |
+
lengths = lengths.to(torch.long)
|
7 |
+
if max_len is None:
|
8 |
+
max_len = torch.max(lengths).item()
|
9 |
+
|
10 |
+
ids = torch.arange(0, max_len).unsqueeze(0).expand(lengths.shape[0], -1).to(lengths.device)
|
11 |
+
mask = ids < lengths.unsqueeze(1).expand(-1, max_len)
|
12 |
+
|
13 |
+
return mask
|
14 |
+
|
15 |
+
|
16 |
+
def linear_interpolation(features, seq_len):
|
17 |
+
features = features.transpose(1, 2)
|
18 |
+
output_features = F.interpolate(features, size=seq_len, align_corners=True, mode='linear')
|
19 |
+
return output_features.transpose(1, 2)
|
20 |
+
|
21 |
+
|
22 |
+
if __name__ == "__main__":
|
23 |
+
import numpy as np
|
24 |
+
mask = ~get_mask_from_lengths(torch.from_numpy(np.array([4,6])))
|
25 |
+
import pdb; pdb.set_trace()
|
src/audio_models/wav2vec2.py
ADDED
@@ -0,0 +1,125 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers import Wav2Vec2Config, Wav2Vec2Model
|
2 |
+
from transformers.modeling_outputs import BaseModelOutput
|
3 |
+
|
4 |
+
from .torch_utils import linear_interpolation
|
5 |
+
|
6 |
+
# the implementation of Wav2Vec2Model is borrowed from
|
7 |
+
# https://github.com/huggingface/transformers/blob/HEAD/src/transformers/models/wav2vec2/modeling_wav2vec2.py
|
8 |
+
# initialize our encoder with the pre-trained wav2vec 2.0 weights.
|
9 |
+
class Wav2Vec2Model(Wav2Vec2Model):
|
10 |
+
def __init__(self, config: Wav2Vec2Config):
|
11 |
+
super().__init__(config)
|
12 |
+
|
13 |
+
def forward(
|
14 |
+
self,
|
15 |
+
input_values,
|
16 |
+
seq_len,
|
17 |
+
attention_mask=None,
|
18 |
+
mask_time_indices=None,
|
19 |
+
output_attentions=None,
|
20 |
+
output_hidden_states=None,
|
21 |
+
return_dict=None,
|
22 |
+
):
|
23 |
+
self.config.output_attentions = True
|
24 |
+
|
25 |
+
output_hidden_states = (
|
26 |
+
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
27 |
+
)
|
28 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
29 |
+
|
30 |
+
extract_features = self.feature_extractor(input_values)
|
31 |
+
extract_features = extract_features.transpose(1, 2)
|
32 |
+
extract_features = linear_interpolation(extract_features, seq_len=seq_len)
|
33 |
+
|
34 |
+
if attention_mask is not None:
|
35 |
+
# compute reduced attention_mask corresponding to feature vectors
|
36 |
+
attention_mask = self._get_feature_vector_attention_mask(
|
37 |
+
extract_features.shape[1], attention_mask, add_adapter=False
|
38 |
+
)
|
39 |
+
|
40 |
+
hidden_states, extract_features = self.feature_projection(extract_features)
|
41 |
+
hidden_states = self._mask_hidden_states(
|
42 |
+
hidden_states, mask_time_indices=mask_time_indices, attention_mask=attention_mask
|
43 |
+
)
|
44 |
+
|
45 |
+
encoder_outputs = self.encoder(
|
46 |
+
hidden_states,
|
47 |
+
attention_mask=attention_mask,
|
48 |
+
output_attentions=output_attentions,
|
49 |
+
output_hidden_states=output_hidden_states,
|
50 |
+
return_dict=return_dict,
|
51 |
+
)
|
52 |
+
|
53 |
+
hidden_states = encoder_outputs[0]
|
54 |
+
|
55 |
+
if self.adapter is not None:
|
56 |
+
hidden_states = self.adapter(hidden_states)
|
57 |
+
|
58 |
+
if not return_dict:
|
59 |
+
return (hidden_states, ) + encoder_outputs[1:]
|
60 |
+
return BaseModelOutput(
|
61 |
+
last_hidden_state=hidden_states,
|
62 |
+
hidden_states=encoder_outputs.hidden_states,
|
63 |
+
attentions=encoder_outputs.attentions,
|
64 |
+
)
|
65 |
+
|
66 |
+
|
67 |
+
def feature_extract(
|
68 |
+
self,
|
69 |
+
input_values,
|
70 |
+
seq_len,
|
71 |
+
):
|
72 |
+
extract_features = self.feature_extractor(input_values)
|
73 |
+
extract_features = extract_features.transpose(1, 2)
|
74 |
+
extract_features = linear_interpolation(extract_features, seq_len=seq_len)
|
75 |
+
|
76 |
+
return extract_features
|
77 |
+
|
78 |
+
def encode(
|
79 |
+
self,
|
80 |
+
extract_features,
|
81 |
+
attention_mask=None,
|
82 |
+
mask_time_indices=None,
|
83 |
+
output_attentions=None,
|
84 |
+
output_hidden_states=None,
|
85 |
+
return_dict=None,
|
86 |
+
):
|
87 |
+
self.config.output_attentions = True
|
88 |
+
|
89 |
+
output_hidden_states = (
|
90 |
+
output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
|
91 |
+
)
|
92 |
+
return_dict = return_dict if return_dict is not None else self.config.use_return_dict
|
93 |
+
|
94 |
+
if attention_mask is not None:
|
95 |
+
# compute reduced attention_mask corresponding to feature vectors
|
96 |
+
attention_mask = self._get_feature_vector_attention_mask(
|
97 |
+
extract_features.shape[1], attention_mask, add_adapter=False
|
98 |
+
)
|
99 |
+
|
100 |
+
|
101 |
+
hidden_states, extract_features = self.feature_projection(extract_features)
|
102 |
+
hidden_states = self._mask_hidden_states(
|
103 |
+
hidden_states, mask_time_indices=mask_time_indices, attention_mask=attention_mask
|
104 |
+
)
|
105 |
+
|
106 |
+
encoder_outputs = self.encoder(
|
107 |
+
hidden_states,
|
108 |
+
attention_mask=attention_mask,
|
109 |
+
output_attentions=output_attentions,
|
110 |
+
output_hidden_states=output_hidden_states,
|
111 |
+
return_dict=return_dict,
|
112 |
+
)
|
113 |
+
|
114 |
+
hidden_states = encoder_outputs[0]
|
115 |
+
|
116 |
+
if self.adapter is not None:
|
117 |
+
hidden_states = self.adapter(hidden_states)
|
118 |
+
|
119 |
+
if not return_dict:
|
120 |
+
return (hidden_states, ) + encoder_outputs[1:]
|
121 |
+
return BaseModelOutput(
|
122 |
+
last_hidden_state=hidden_states,
|
123 |
+
hidden_states=encoder_outputs.hidden_states,
|
124 |
+
attentions=encoder_outputs.attentions,
|
125 |
+
)
|
src/models/attention.py
ADDED
@@ -0,0 +1,840 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention.py
|
2 |
+
|
3 |
+
from typing import Any, Dict, Optional
|
4 |
+
|
5 |
+
import torch
|
6 |
+
from diffusers.models.attention import AdaLayerNorm, Attention, FeedForward
|
7 |
+
from diffusers.models.embeddings import SinusoidalPositionalEmbedding
|
8 |
+
from einops import rearrange
|
9 |
+
from torch import nn
|
10 |
+
|
11 |
+
from diffusers.models.attention import *
|
12 |
+
from diffusers.models.attention_processor import *
|
13 |
+
|
14 |
+
class BasicTransformerBlock(nn.Module):
|
15 |
+
r"""
|
16 |
+
A basic Transformer block.
|
17 |
+
|
18 |
+
Parameters:
|
19 |
+
dim (`int`): The number of channels in the input and output.
|
20 |
+
num_attention_heads (`int`): The number of heads to use for multi-head attention.
|
21 |
+
attention_head_dim (`int`): The number of channels in each head.
|
22 |
+
dropout (`float`, *optional*, defaults to 0.0): The dropout probability to use.
|
23 |
+
cross_attention_dim (`int`, *optional*): The size of the encoder_hidden_states vector for cross attention.
|
24 |
+
activation_fn (`str`, *optional*, defaults to `"geglu"`): Activation function to be used in feed-forward.
|
25 |
+
num_embeds_ada_norm (:
|
26 |
+
obj: `int`, *optional*): The number of diffusion steps used during training. See `Transformer2DModel`.
|
27 |
+
attention_bias (:
|
28 |
+
obj: `bool`, *optional*, defaults to `False`): Configure if the attentions should contain a bias parameter.
|
29 |
+
only_cross_attention (`bool`, *optional*):
|
30 |
+
Whether to use only cross-attention layers. In this case two cross attention layers are used.
|
31 |
+
double_self_attention (`bool`, *optional*):
|
32 |
+
Whether to use two self-attention layers. In this case no cross attention layers are used.
|
33 |
+
upcast_attention (`bool`, *optional*):
|
34 |
+
Whether to upcast the attention computation to float32. This is useful for mixed precision training.
|
35 |
+
norm_elementwise_affine (`bool`, *optional*, defaults to `True`):
|
36 |
+
Whether to use learnable elementwise affine parameters for normalization.
|
37 |
+
norm_type (`str`, *optional*, defaults to `"layer_norm"`):
|
38 |
+
The normalization layer to use. Can be `"layer_norm"`, `"ada_norm"` or `"ada_norm_zero"`.
|
39 |
+
final_dropout (`bool` *optional*, defaults to False):
|
40 |
+
Whether to apply a final dropout after the last feed-forward layer.
|
41 |
+
attention_type (`str`, *optional*, defaults to `"default"`):
|
42 |
+
The type of attention to use. Can be `"default"` or `"gated"` or `"gated-text-image"`.
|
43 |
+
positional_embeddings (`str`, *optional*, defaults to `None`):
|
44 |
+
The type of positional embeddings to apply to.
|
45 |
+
num_positional_embeddings (`int`, *optional*, defaults to `None`):
|
46 |
+
The maximum number of positional embeddings to apply.
|
47 |
+
"""
|
48 |
+
|
49 |
+
def __init__(
|
50 |
+
self,
|
51 |
+
dim: int,
|
52 |
+
num_attention_heads: int,
|
53 |
+
attention_head_dim: int,
|
54 |
+
dropout=0.0,
|
55 |
+
cross_attention_dim: Optional[int] = None,
|
56 |
+
activation_fn: str = "geglu",
|
57 |
+
num_embeds_ada_norm: Optional[int] = None,
|
58 |
+
attention_bias: bool = False,
|
59 |
+
only_cross_attention: bool = False,
|
60 |
+
double_self_attention: bool = False,
|
61 |
+
upcast_attention: bool = False,
|
62 |
+
norm_elementwise_affine: bool = True,
|
63 |
+
norm_type: str = "layer_norm", # 'layer_norm', 'ada_norm', 'ada_norm_zero', 'ada_norm_single'
|
64 |
+
norm_eps: float = 1e-5,
|
65 |
+
final_dropout: bool = False,
|
66 |
+
attention_type: str = "default",
|
67 |
+
positional_embeddings: Optional[str] = None,
|
68 |
+
num_positional_embeddings: Optional[int] = None,
|
69 |
+
):
|
70 |
+
super().__init__()
|
71 |
+
self.only_cross_attention = only_cross_attention
|
72 |
+
|
73 |
+
self.use_ada_layer_norm_zero = (
|
74 |
+
num_embeds_ada_norm is not None
|
75 |
+
) and norm_type == "ada_norm_zero"
|
76 |
+
self.use_ada_layer_norm = (
|
77 |
+
num_embeds_ada_norm is not None
|
78 |
+
) and norm_type == "ada_norm"
|
79 |
+
self.use_ada_layer_norm_single = norm_type == "ada_norm_single"
|
80 |
+
self.use_layer_norm = norm_type == "layer_norm"
|
81 |
+
|
82 |
+
if norm_type in ("ada_norm", "ada_norm_zero") and num_embeds_ada_norm is None:
|
83 |
+
raise ValueError(
|
84 |
+
f"`norm_type` is set to {norm_type}, but `num_embeds_ada_norm` is not defined. Please make sure to"
|
85 |
+
f" define `num_embeds_ada_norm` if setting `norm_type` to {norm_type}."
|
86 |
+
)
|
87 |
+
|
88 |
+
if positional_embeddings and (num_positional_embeddings is None):
|
89 |
+
raise ValueError(
|
90 |
+
"If `positional_embedding` type is defined, `num_positition_embeddings` must also be defined."
|
91 |
+
)
|
92 |
+
|
93 |
+
if positional_embeddings == "sinusoidal":
|
94 |
+
self.pos_embed = SinusoidalPositionalEmbedding(
|
95 |
+
dim, max_seq_length=num_positional_embeddings
|
96 |
+
)
|
97 |
+
else:
|
98 |
+
self.pos_embed = None
|
99 |
+
|
100 |
+
# Define 3 blocks. Each block has its own normalization layer.
|
101 |
+
# 1. Self-Attn
|
102 |
+
if self.use_ada_layer_norm:
|
103 |
+
self.norm1 = AdaLayerNorm(dim, num_embeds_ada_norm)
|
104 |
+
elif self.use_ada_layer_norm_zero:
|
105 |
+
self.norm1 = AdaLayerNormZero(dim, num_embeds_ada_norm)
|
106 |
+
else:
|
107 |
+
self.norm1 = nn.LayerNorm(
|
108 |
+
dim, elementwise_affine=norm_elementwise_affine, eps=norm_eps
|
109 |
+
)
|
110 |
+
|
111 |
+
self.attn1 = Attention(
|
112 |
+
query_dim=dim,
|
113 |
+
heads=num_attention_heads,
|
114 |
+
dim_head=attention_head_dim,
|
115 |
+
dropout=dropout,
|
116 |
+
bias=attention_bias,
|
117 |
+
cross_attention_dim=cross_attention_dim if only_cross_attention else None,
|
118 |
+
upcast_attention=upcast_attention,
|
119 |
+
)
|
120 |
+
|
121 |
+
# 2. Cross-Attn
|
122 |
+
if cross_attention_dim is not None or double_self_attention:
|
123 |
+
# We currently only use AdaLayerNormZero for self attention where there will only be one attention block.
|
124 |
+
# I.e. the number of returned modulation chunks from AdaLayerZero would not make sense if returned during
|
125 |
+
# the second cross attention block.
|
126 |
+
self.norm2 = (
|
127 |
+
AdaLayerNorm(dim, num_embeds_ada_norm)
|
128 |
+
if self.use_ada_layer_norm
|
129 |
+
else nn.LayerNorm(
|
130 |
+
dim, elementwise_affine=norm_elementwise_affine, eps=norm_eps
|
131 |
+
)
|
132 |
+
)
|
133 |
+
self.attn2 = Attention(
|
134 |
+
query_dim=dim,
|
135 |
+
cross_attention_dim=cross_attention_dim
|
136 |
+
if not double_self_attention
|
137 |
+
else None,
|
138 |
+
heads=num_attention_heads,
|
139 |
+
dim_head=attention_head_dim,
|
140 |
+
dropout=dropout,
|
141 |
+
bias=attention_bias,
|
142 |
+
upcast_attention=upcast_attention,
|
143 |
+
) # is self-attn if encoder_hidden_states is none
|
144 |
+
else:
|
145 |
+
self.norm2 = None
|
146 |
+
self.attn2 = None
|
147 |
+
|
148 |
+
# 3. Feed-forward
|
149 |
+
if not self.use_ada_layer_norm_single:
|
150 |
+
self.norm3 = nn.LayerNorm(
|
151 |
+
dim, elementwise_affine=norm_elementwise_affine, eps=norm_eps
|
152 |
+
)
|
153 |
+
|
154 |
+
self.ff = FeedForward(
|
155 |
+
dim,
|
156 |
+
dropout=dropout,
|
157 |
+
activation_fn=activation_fn,
|
158 |
+
final_dropout=final_dropout,
|
159 |
+
)
|
160 |
+
|
161 |
+
# 4. Fuser
|
162 |
+
if attention_type == "gated" or attention_type == "gated-text-image":
|
163 |
+
self.fuser = GatedSelfAttentionDense(
|
164 |
+
dim, cross_attention_dim, num_attention_heads, attention_head_dim
|
165 |
+
)
|
166 |
+
|
167 |
+
# 5. Scale-shift for PixArt-Alpha.
|
168 |
+
if self.use_ada_layer_norm_single:
|
169 |
+
self.scale_shift_table = nn.Parameter(torch.randn(6, dim) / dim**0.5)
|
170 |
+
|
171 |
+
# let chunk size default to None
|
172 |
+
self._chunk_size = None
|
173 |
+
self._chunk_dim = 0
|
174 |
+
|
175 |
+
def set_chunk_feed_forward(self, chunk_size: Optional[int], dim: int = 0):
|
176 |
+
# Sets chunk feed-forward
|
177 |
+
self._chunk_size = chunk_size
|
178 |
+
self._chunk_dim = dim
|
179 |
+
|
180 |
+
def forward(
|
181 |
+
self,
|
182 |
+
hidden_states: torch.FloatTensor,
|
183 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
184 |
+
encoder_hidden_states: Optional[torch.FloatTensor] = None,
|
185 |
+
encoder_attention_mask: Optional[torch.FloatTensor] = None,
|
186 |
+
timestep: Optional[torch.LongTensor] = None,
|
187 |
+
cross_attention_kwargs: Dict[str, Any] = None,
|
188 |
+
class_labels: Optional[torch.LongTensor] = None,
|
189 |
+
) -> torch.FloatTensor:
|
190 |
+
# Notice that normalization is always applied before the real computation in the following blocks.
|
191 |
+
# 0. Self-Attention
|
192 |
+
batch_size = hidden_states.shape[0]
|
193 |
+
|
194 |
+
if self.use_ada_layer_norm:
|
195 |
+
norm_hidden_states = self.norm1(hidden_states, timestep)
|
196 |
+
elif self.use_ada_layer_norm_zero:
|
197 |
+
norm_hidden_states, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.norm1(
|
198 |
+
hidden_states, timestep, class_labels, hidden_dtype=hidden_states.dtype
|
199 |
+
)
|
200 |
+
elif self.use_layer_norm:
|
201 |
+
norm_hidden_states = self.norm1(hidden_states)
|
202 |
+
elif self.use_ada_layer_norm_single:
|
203 |
+
shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = (
|
204 |
+
self.scale_shift_table[None] + timestep.reshape(batch_size, 6, -1)
|
205 |
+
).chunk(6, dim=1)
|
206 |
+
norm_hidden_states = self.norm1(hidden_states)
|
207 |
+
norm_hidden_states = norm_hidden_states * (1 + scale_msa) + shift_msa
|
208 |
+
norm_hidden_states = norm_hidden_states.squeeze(1)
|
209 |
+
else:
|
210 |
+
raise ValueError("Incorrect norm used")
|
211 |
+
|
212 |
+
if self.pos_embed is not None:
|
213 |
+
norm_hidden_states = self.pos_embed(norm_hidden_states)
|
214 |
+
|
215 |
+
# 1. Retrieve lora scale.
|
216 |
+
lora_scale = (
|
217 |
+
cross_attention_kwargs.get("scale", 1.0)
|
218 |
+
if cross_attention_kwargs is not None
|
219 |
+
else 1.0
|
220 |
+
)
|
221 |
+
|
222 |
+
# 2. Prepare GLIGEN inputs
|
223 |
+
cross_attention_kwargs = (
|
224 |
+
cross_attention_kwargs.copy() if cross_attention_kwargs is not None else {}
|
225 |
+
)
|
226 |
+
gligen_kwargs = cross_attention_kwargs.pop("gligen", None)
|
227 |
+
|
228 |
+
attn_output = self.attn1(
|
229 |
+
norm_hidden_states,
|
230 |
+
encoder_hidden_states=encoder_hidden_states
|
231 |
+
if self.only_cross_attention
|
232 |
+
else None,
|
233 |
+
attention_mask=attention_mask,
|
234 |
+
**cross_attention_kwargs,
|
235 |
+
)
|
236 |
+
if self.use_ada_layer_norm_zero:
|
237 |
+
attn_output = gate_msa.unsqueeze(1) * attn_output
|
238 |
+
elif self.use_ada_layer_norm_single:
|
239 |
+
attn_output = gate_msa * attn_output
|
240 |
+
|
241 |
+
hidden_states = attn_output + hidden_states
|
242 |
+
if hidden_states.ndim == 4:
|
243 |
+
hidden_states = hidden_states.squeeze(1)
|
244 |
+
|
245 |
+
# 2.5 GLIGEN Control
|
246 |
+
if gligen_kwargs is not None:
|
247 |
+
hidden_states = self.fuser(hidden_states, gligen_kwargs["objs"])
|
248 |
+
|
249 |
+
# 3. Cross-Attention
|
250 |
+
if self.attn2 is not None:
|
251 |
+
if self.use_ada_layer_norm:
|
252 |
+
norm_hidden_states = self.norm2(hidden_states, timestep)
|
253 |
+
elif self.use_ada_layer_norm_zero or self.use_layer_norm:
|
254 |
+
norm_hidden_states = self.norm2(hidden_states)
|
255 |
+
elif self.use_ada_layer_norm_single:
|
256 |
+
# For PixArt norm2 isn't applied here:
|
257 |
+
# https://github.com/PixArt-alpha/PixArt-alpha/blob/0f55e922376d8b797edd44d25d0e7464b260dcab/diffusion/model/nets/PixArtMS.py#L70C1-L76C103
|
258 |
+
norm_hidden_states = hidden_states
|
259 |
+
else:
|
260 |
+
raise ValueError("Incorrect norm")
|
261 |
+
|
262 |
+
if self.pos_embed is not None and self.use_ada_layer_norm_single is False:
|
263 |
+
norm_hidden_states = self.pos_embed(norm_hidden_states)
|
264 |
+
|
265 |
+
attn_output = self.attn2(
|
266 |
+
norm_hidden_states,
|
267 |
+
encoder_hidden_states=encoder_hidden_states,
|
268 |
+
attention_mask=encoder_attention_mask,
|
269 |
+
**cross_attention_kwargs,
|
270 |
+
)
|
271 |
+
hidden_states = attn_output + hidden_states
|
272 |
+
|
273 |
+
# 4. Feed-forward
|
274 |
+
if not self.use_ada_layer_norm_single:
|
275 |
+
norm_hidden_states = self.norm3(hidden_states)
|
276 |
+
|
277 |
+
if self.use_ada_layer_norm_zero:
|
278 |
+
norm_hidden_states = (
|
279 |
+
norm_hidden_states * (1 + scale_mlp[:, None]) + shift_mlp[:, None]
|
280 |
+
)
|
281 |
+
|
282 |
+
if self.use_ada_layer_norm_single:
|
283 |
+
norm_hidden_states = self.norm2(hidden_states)
|
284 |
+
norm_hidden_states = norm_hidden_states * (1 + scale_mlp) + shift_mlp
|
285 |
+
|
286 |
+
ff_output = self.ff(norm_hidden_states, scale=lora_scale)
|
287 |
+
|
288 |
+
if self.use_ada_layer_norm_zero:
|
289 |
+
ff_output = gate_mlp.unsqueeze(1) * ff_output
|
290 |
+
elif self.use_ada_layer_norm_single:
|
291 |
+
ff_output = gate_mlp * ff_output
|
292 |
+
|
293 |
+
hidden_states = ff_output + hidden_states
|
294 |
+
if hidden_states.ndim == 4:
|
295 |
+
hidden_states = hidden_states.squeeze(1)
|
296 |
+
|
297 |
+
return hidden_states
|
298 |
+
|
299 |
+
|
300 |
+
class TemporalBasicTransformerBlock(nn.Module):
|
301 |
+
def __init__(
|
302 |
+
self,
|
303 |
+
dim: int,
|
304 |
+
num_attention_heads: int,
|
305 |
+
attention_head_dim: int,
|
306 |
+
dropout=0.0,
|
307 |
+
cross_attention_dim: Optional[int] = None,
|
308 |
+
activation_fn: str = "geglu",
|
309 |
+
num_embeds_ada_norm: Optional[int] = None,
|
310 |
+
attention_bias: bool = False,
|
311 |
+
only_cross_attention: bool = False,
|
312 |
+
upcast_attention: bool = False,
|
313 |
+
unet_use_cross_frame_attention=None,
|
314 |
+
unet_use_temporal_attention=None,
|
315 |
+
):
|
316 |
+
super().__init__()
|
317 |
+
self.only_cross_attention = only_cross_attention
|
318 |
+
self.use_ada_layer_norm = num_embeds_ada_norm is not None
|
319 |
+
self.unet_use_cross_frame_attention = unet_use_cross_frame_attention
|
320 |
+
self.unet_use_temporal_attention = unet_use_temporal_attention
|
321 |
+
|
322 |
+
# SC-Attn
|
323 |
+
self.attn1 = Attention(
|
324 |
+
query_dim=dim,
|
325 |
+
heads=num_attention_heads,
|
326 |
+
dim_head=attention_head_dim,
|
327 |
+
dropout=dropout,
|
328 |
+
bias=attention_bias,
|
329 |
+
upcast_attention=upcast_attention,
|
330 |
+
)
|
331 |
+
self.norm1 = (
|
332 |
+
AdaLayerNorm(dim, num_embeds_ada_norm)
|
333 |
+
if self.use_ada_layer_norm
|
334 |
+
else nn.LayerNorm(dim)
|
335 |
+
)
|
336 |
+
|
337 |
+
# Cross-Attn
|
338 |
+
if cross_attention_dim is not None:
|
339 |
+
self.attn2 = Attention(
|
340 |
+
query_dim=dim,
|
341 |
+
cross_attention_dim=cross_attention_dim,
|
342 |
+
heads=num_attention_heads,
|
343 |
+
dim_head=attention_head_dim,
|
344 |
+
dropout=dropout,
|
345 |
+
bias=attention_bias,
|
346 |
+
upcast_attention=upcast_attention,
|
347 |
+
)
|
348 |
+
else:
|
349 |
+
self.attn2 = None
|
350 |
+
|
351 |
+
if cross_attention_dim is not None:
|
352 |
+
self.norm2 = (
|
353 |
+
AdaLayerNorm(dim, num_embeds_ada_norm)
|
354 |
+
if self.use_ada_layer_norm
|
355 |
+
else nn.LayerNorm(dim)
|
356 |
+
)
|
357 |
+
else:
|
358 |
+
self.norm2 = None
|
359 |
+
|
360 |
+
# Feed-forward
|
361 |
+
self.ff = FeedForward(dim, dropout=dropout, activation_fn=activation_fn)
|
362 |
+
self.norm3 = nn.LayerNorm(dim)
|
363 |
+
self.use_ada_layer_norm_zero = False
|
364 |
+
|
365 |
+
# Temp-Attn
|
366 |
+
assert unet_use_temporal_attention is not None
|
367 |
+
if unet_use_temporal_attention:
|
368 |
+
self.attn_temp = Attention(
|
369 |
+
query_dim=dim,
|
370 |
+
heads=num_attention_heads,
|
371 |
+
dim_head=attention_head_dim,
|
372 |
+
dropout=dropout,
|
373 |
+
bias=attention_bias,
|
374 |
+
upcast_attention=upcast_attention,
|
375 |
+
)
|
376 |
+
nn.init.zeros_(self.attn_temp.to_out[0].weight.data)
|
377 |
+
self.norm_temp = (
|
378 |
+
AdaLayerNorm(dim, num_embeds_ada_norm)
|
379 |
+
if self.use_ada_layer_norm
|
380 |
+
else nn.LayerNorm(dim)
|
381 |
+
)
|
382 |
+
|
383 |
+
def forward(
|
384 |
+
self,
|
385 |
+
hidden_states,
|
386 |
+
encoder_hidden_states=None,
|
387 |
+
timestep=None,
|
388 |
+
attention_mask=None,
|
389 |
+
video_length=None,
|
390 |
+
):
|
391 |
+
norm_hidden_states = (
|
392 |
+
self.norm1(hidden_states, timestep)
|
393 |
+
if self.use_ada_layer_norm
|
394 |
+
else self.norm1(hidden_states)
|
395 |
+
)
|
396 |
+
|
397 |
+
if self.unet_use_cross_frame_attention:
|
398 |
+
hidden_states = (
|
399 |
+
self.attn1(
|
400 |
+
norm_hidden_states,
|
401 |
+
attention_mask=attention_mask,
|
402 |
+
video_length=video_length,
|
403 |
+
)
|
404 |
+
+ hidden_states
|
405 |
+
)
|
406 |
+
else:
|
407 |
+
hidden_states = (
|
408 |
+
self.attn1(norm_hidden_states, attention_mask=attention_mask)
|
409 |
+
+ hidden_states
|
410 |
+
)
|
411 |
+
|
412 |
+
if self.attn2 is not None:
|
413 |
+
# Cross-Attention
|
414 |
+
norm_hidden_states = (
|
415 |
+
self.norm2(hidden_states, timestep)
|
416 |
+
if self.use_ada_layer_norm
|
417 |
+
else self.norm2(hidden_states)
|
418 |
+
)
|
419 |
+
hidden_states = (
|
420 |
+
self.attn2(
|
421 |
+
norm_hidden_states,
|
422 |
+
encoder_hidden_states=encoder_hidden_states,
|
423 |
+
attention_mask=attention_mask,
|
424 |
+
)
|
425 |
+
+ hidden_states
|
426 |
+
)
|
427 |
+
|
428 |
+
# Feed-forward
|
429 |
+
hidden_states = self.ff(self.norm3(hidden_states)) + hidden_states
|
430 |
+
|
431 |
+
# Temporal-Attention
|
432 |
+
if self.unet_use_temporal_attention:
|
433 |
+
d = hidden_states.shape[1]
|
434 |
+
hidden_states = rearrange(
|
435 |
+
hidden_states, "(b f) d c -> (b d) f c", f=video_length
|
436 |
+
)
|
437 |
+
norm_hidden_states = (
|
438 |
+
self.norm_temp(hidden_states, timestep)
|
439 |
+
if self.use_ada_layer_norm
|
440 |
+
else self.norm_temp(hidden_states)
|
441 |
+
)
|
442 |
+
hidden_states = self.attn_temp(norm_hidden_states) + hidden_states
|
443 |
+
hidden_states = rearrange(hidden_states, "(b d) f c -> (b f) d c", d=d)
|
444 |
+
|
445 |
+
return hidden_states
|
446 |
+
|
447 |
+
|
448 |
+
class ResidualTemporalBasicTransformerBlock(TemporalBasicTransformerBlock):
|
449 |
+
def __init__(
|
450 |
+
self,
|
451 |
+
dim: int,
|
452 |
+
num_attention_heads: int,
|
453 |
+
attention_head_dim: int,
|
454 |
+
dropout=0.0,
|
455 |
+
cross_attention_dim: Optional[int] = None,
|
456 |
+
activation_fn: str = "geglu",
|
457 |
+
num_embeds_ada_norm: Optional[int] = None,
|
458 |
+
attention_bias: bool = False,
|
459 |
+
only_cross_attention: bool = False,
|
460 |
+
upcast_attention: bool = False,
|
461 |
+
unet_use_cross_frame_attention=None,
|
462 |
+
unet_use_temporal_attention=None,
|
463 |
+
):
|
464 |
+
super(TemporalBasicTransformerBlock, self).__init__()
|
465 |
+
self.only_cross_attention = only_cross_attention
|
466 |
+
self.use_ada_layer_norm = num_embeds_ada_norm is not None
|
467 |
+
self.unet_use_cross_frame_attention = unet_use_cross_frame_attention
|
468 |
+
self.unet_use_temporal_attention = unet_use_temporal_attention
|
469 |
+
|
470 |
+
# SC-Attn
|
471 |
+
self.attn1 = ResidualAttention(
|
472 |
+
query_dim=dim,
|
473 |
+
heads=num_attention_heads,
|
474 |
+
dim_head=attention_head_dim,
|
475 |
+
dropout=dropout,
|
476 |
+
bias=attention_bias,
|
477 |
+
upcast_attention=upcast_attention,
|
478 |
+
)
|
479 |
+
self.norm1 = (
|
480 |
+
AdaLayerNorm(dim, num_embeds_ada_norm)
|
481 |
+
if self.use_ada_layer_norm
|
482 |
+
else nn.LayerNorm(dim)
|
483 |
+
)
|
484 |
+
|
485 |
+
# Cross-Attn
|
486 |
+
if cross_attention_dim is not None:
|
487 |
+
self.attn2 = ResidualAttention(
|
488 |
+
query_dim=dim,
|
489 |
+
cross_attention_dim=cross_attention_dim,
|
490 |
+
heads=num_attention_heads,
|
491 |
+
dim_head=attention_head_dim,
|
492 |
+
dropout=dropout,
|
493 |
+
bias=attention_bias,
|
494 |
+
upcast_attention=upcast_attention,
|
495 |
+
)
|
496 |
+
else:
|
497 |
+
self.attn2 = None
|
498 |
+
|
499 |
+
if cross_attention_dim is not None:
|
500 |
+
self.norm2 = (
|
501 |
+
AdaLayerNorm(dim, num_embeds_ada_norm)
|
502 |
+
if self.use_ada_layer_norm
|
503 |
+
else nn.LayerNorm(dim)
|
504 |
+
)
|
505 |
+
else:
|
506 |
+
self.norm2 = None
|
507 |
+
|
508 |
+
# Feed-forward
|
509 |
+
self.ff = FeedForward(dim, dropout=dropout, activation_fn=activation_fn)
|
510 |
+
self.norm3 = nn.LayerNorm(dim)
|
511 |
+
self.use_ada_layer_norm_zero = False
|
512 |
+
|
513 |
+
# Temp-Attn
|
514 |
+
assert unet_use_temporal_attention is not None
|
515 |
+
if unet_use_temporal_attention:
|
516 |
+
self.attn_temp = Attention(
|
517 |
+
query_dim=dim,
|
518 |
+
heads=num_attention_heads,
|
519 |
+
dim_head=attention_head_dim,
|
520 |
+
dropout=dropout,
|
521 |
+
bias=attention_bias,
|
522 |
+
upcast_attention=upcast_attention,
|
523 |
+
)
|
524 |
+
nn.init.zeros_(self.attn_temp.to_out[0].weight.data)
|
525 |
+
self.norm_temp = (
|
526 |
+
AdaLayerNorm(dim, num_embeds_ada_norm)
|
527 |
+
if self.use_ada_layer_norm
|
528 |
+
else nn.LayerNorm(dim)
|
529 |
+
)
|
530 |
+
|
531 |
+
def forward(
|
532 |
+
self,
|
533 |
+
hidden_states,
|
534 |
+
encoder_hidden_states=None,
|
535 |
+
timestep=None,
|
536 |
+
attention_mask=None,
|
537 |
+
video_length=None,
|
538 |
+
block_idx: Optional[int] = None,
|
539 |
+
additional_residuals: Optional[Dict[str, torch.FloatTensor]] = None
|
540 |
+
):
|
541 |
+
norm_hidden_states = (
|
542 |
+
self.norm1(hidden_states, timestep)
|
543 |
+
if self.use_ada_layer_norm
|
544 |
+
else self.norm1(hidden_states)
|
545 |
+
)
|
546 |
+
|
547 |
+
if self.unet_use_cross_frame_attention:
|
548 |
+
hidden_states = (
|
549 |
+
self.attn1(
|
550 |
+
norm_hidden_states,
|
551 |
+
attention_mask=attention_mask,
|
552 |
+
video_length=video_length,
|
553 |
+
block_idx=block_idx,
|
554 |
+
additional_residuals=additional_residuals,
|
555 |
+
)
|
556 |
+
+ hidden_states
|
557 |
+
)
|
558 |
+
else:
|
559 |
+
hidden_states = (
|
560 |
+
self.attn1(norm_hidden_states, attention_mask=attention_mask,
|
561 |
+
block_idx=block_idx,
|
562 |
+
additional_residuals=additional_residuals
|
563 |
+
)
|
564 |
+
+ hidden_states
|
565 |
+
)
|
566 |
+
|
567 |
+
if self.attn2 is not None:
|
568 |
+
# Cross-Attention
|
569 |
+
norm_hidden_states = (
|
570 |
+
self.norm2(hidden_states, timestep)
|
571 |
+
if self.use_ada_layer_norm
|
572 |
+
else self.norm2(hidden_states)
|
573 |
+
)
|
574 |
+
hidden_states = (
|
575 |
+
self.attn2(
|
576 |
+
norm_hidden_states,
|
577 |
+
encoder_hidden_states=encoder_hidden_states,
|
578 |
+
attention_mask=attention_mask,
|
579 |
+
block_idx=block_idx,
|
580 |
+
additional_residuals=additional_residuals,
|
581 |
+
)
|
582 |
+
+ hidden_states
|
583 |
+
)
|
584 |
+
|
585 |
+
# Feed-forward
|
586 |
+
hidden_states = self.ff(self.norm3(hidden_states)) + hidden_states
|
587 |
+
|
588 |
+
# Temporal-Attention
|
589 |
+
if self.unet_use_temporal_attention:
|
590 |
+
d = hidden_states.shape[1]
|
591 |
+
hidden_states = rearrange(
|
592 |
+
hidden_states, "(b f) d c -> (b d) f c", f=video_length
|
593 |
+
)
|
594 |
+
norm_hidden_states = (
|
595 |
+
self.norm_temp(hidden_states, timestep)
|
596 |
+
if self.use_ada_layer_norm
|
597 |
+
else self.norm_temp(hidden_states)
|
598 |
+
)
|
599 |
+
hidden_states = self.attn_temp(norm_hidden_states) + hidden_states
|
600 |
+
hidden_states = rearrange(hidden_states, "(b d) f c -> (b f) d c", d=d)
|
601 |
+
|
602 |
+
return hidden_states
|
603 |
+
|
604 |
+
|
605 |
+
class ResidualAttention(Attention):
|
606 |
+
def set_use_memory_efficient_attention_xformers(
|
607 |
+
self, use_memory_efficient_attention_xformers: bool, attention_op: Optional[Callable] = None
|
608 |
+
):
|
609 |
+
is_lora = hasattr(self, "processor") and isinstance(
|
610 |
+
self.processor,
|
611 |
+
LORA_ATTENTION_PROCESSORS,
|
612 |
+
)
|
613 |
+
is_custom_diffusion = hasattr(self, "processor") and isinstance(
|
614 |
+
self.processor, (CustomDiffusionAttnProcessor, CustomDiffusionXFormersAttnProcessor)
|
615 |
+
)
|
616 |
+
is_added_kv_processor = hasattr(self, "processor") and isinstance(
|
617 |
+
self.processor,
|
618 |
+
(
|
619 |
+
AttnAddedKVProcessor,
|
620 |
+
AttnAddedKVProcessor2_0,
|
621 |
+
SlicedAttnAddedKVProcessor,
|
622 |
+
XFormersAttnAddedKVProcessor,
|
623 |
+
LoRAAttnAddedKVProcessor,
|
624 |
+
),
|
625 |
+
)
|
626 |
+
|
627 |
+
if use_memory_efficient_attention_xformers:
|
628 |
+
if is_added_kv_processor and (is_lora or is_custom_diffusion):
|
629 |
+
raise NotImplementedError(
|
630 |
+
f"Memory efficient attention is currently not supported for LoRA or custom diffuson for attention processor type {self.processor}"
|
631 |
+
)
|
632 |
+
if not is_xformers_available():
|
633 |
+
raise ModuleNotFoundError(
|
634 |
+
(
|
635 |
+
"Refer to https://github.com/facebookresearch/xformers for more information on how to install"
|
636 |
+
" xformers"
|
637 |
+
),
|
638 |
+
name="xformers",
|
639 |
+
)
|
640 |
+
elif not torch.cuda.is_available():
|
641 |
+
raise ValueError(
|
642 |
+
"torch.cuda.is_available() should be True but is False. xformers' memory efficient attention is"
|
643 |
+
" only available for GPU "
|
644 |
+
)
|
645 |
+
else:
|
646 |
+
try:
|
647 |
+
# Make sure we can run the memory efficient attention
|
648 |
+
_ = xformers.ops.memory_efficient_attention(
|
649 |
+
torch.randn((1, 2, 40), device="cuda"),
|
650 |
+
torch.randn((1, 2, 40), device="cuda"),
|
651 |
+
torch.randn((1, 2, 40), device="cuda"),
|
652 |
+
)
|
653 |
+
except Exception as e:
|
654 |
+
raise e
|
655 |
+
|
656 |
+
if is_lora:
|
657 |
+
# TODO (sayakpaul): should we throw a warning if someone wants to use the xformers
|
658 |
+
# variant when using PT 2.0 now that we have LoRAAttnProcessor2_0?
|
659 |
+
processor = LoRAXFormersAttnProcessor(
|
660 |
+
hidden_size=self.processor.hidden_size,
|
661 |
+
cross_attention_dim=self.processor.cross_attention_dim,
|
662 |
+
rank=self.processor.rank,
|
663 |
+
attention_op=attention_op,
|
664 |
+
)
|
665 |
+
processor.load_state_dict(self.processor.state_dict())
|
666 |
+
processor.to(self.processor.to_q_lora.up.weight.device)
|
667 |
+
elif is_custom_diffusion:
|
668 |
+
processor = CustomDiffusionXFormersAttnProcessor(
|
669 |
+
train_kv=self.processor.train_kv,
|
670 |
+
train_q_out=self.processor.train_q_out,
|
671 |
+
hidden_size=self.processor.hidden_size,
|
672 |
+
cross_attention_dim=self.processor.cross_attention_dim,
|
673 |
+
attention_op=attention_op,
|
674 |
+
)
|
675 |
+
processor.load_state_dict(self.processor.state_dict())
|
676 |
+
if hasattr(self.processor, "to_k_custom_diffusion"):
|
677 |
+
processor.to(self.processor.to_k_custom_diffusion.weight.device)
|
678 |
+
elif is_added_kv_processor:
|
679 |
+
# TODO(Patrick, Suraj, William) - currently xformers doesn't work for UnCLIP
|
680 |
+
# which uses this type of cross attention ONLY because the attention mask of format
|
681 |
+
# [0, ..., -10.000, ..., 0, ...,] is not supported
|
682 |
+
# throw warning
|
683 |
+
logger.info(
|
684 |
+
"Memory efficient attention with `xformers` might currently not work correctly if an attention mask is required for the attention operation."
|
685 |
+
)
|
686 |
+
processor = XFormersAttnAddedKVProcessor(attention_op=attention_op)
|
687 |
+
else:
|
688 |
+
processor = ResidualXFormersAttnProcessor(attention_op=attention_op)
|
689 |
+
else:
|
690 |
+
if is_lora:
|
691 |
+
attn_processor_class = (
|
692 |
+
LoRAAttnProcessor2_0 if hasattr(F, "scaled_dot_product_attention") else LoRAAttnProcessor
|
693 |
+
)
|
694 |
+
processor = attn_processor_class(
|
695 |
+
hidden_size=self.processor.hidden_size,
|
696 |
+
cross_attention_dim=self.processor.cross_attention_dim,
|
697 |
+
rank=self.processor.rank,
|
698 |
+
)
|
699 |
+
processor.load_state_dict(self.processor.state_dict())
|
700 |
+
processor.to(self.processor.to_q_lora.up.weight.device)
|
701 |
+
elif is_custom_diffusion:
|
702 |
+
processor = CustomDiffusionAttnProcessor(
|
703 |
+
train_kv=self.processor.train_kv,
|
704 |
+
train_q_out=self.processor.train_q_out,
|
705 |
+
hidden_size=self.processor.hidden_size,
|
706 |
+
cross_attention_dim=self.processor.cross_attention_dim,
|
707 |
+
)
|
708 |
+
processor.load_state_dict(self.processor.state_dict())
|
709 |
+
if hasattr(self.processor, "to_k_custom_diffusion"):
|
710 |
+
processor.to(self.processor.to_k_custom_diffusion.weight.device)
|
711 |
+
else:
|
712 |
+
# set attention processor
|
713 |
+
# We use the AttnProcessor2_0 by default when torch 2.x is used which uses
|
714 |
+
# torch.nn.functional.scaled_dot_product_attention for native Flash/memory_efficient_attention
|
715 |
+
# but only if it has the default `scale` argument. TODO remove scale_qk check when we move to torch 2.1
|
716 |
+
processor = (
|
717 |
+
AttnProcessor2_0()
|
718 |
+
if hasattr(F, "scaled_dot_product_attention") and self.scale_qk
|
719 |
+
else AttnProcessor()
|
720 |
+
)
|
721 |
+
|
722 |
+
self.set_processor(processor)
|
723 |
+
|
724 |
+
def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None,
|
725 |
+
block_idx: Optional[int] = None, additional_residuals: Optional[Dict[str, torch.FloatTensor]] = None,
|
726 |
+
is_self_attn: Optional[bool] = None, **cross_attention_kwargs):
|
727 |
+
# The `Attention` class can call different attention processors / attention functions
|
728 |
+
# here we simply pass along all tensors to the selected processor class
|
729 |
+
# For standard processors that are defined here, `**cross_attention_kwargs` is empty
|
730 |
+
return self.processor(
|
731 |
+
self,
|
732 |
+
hidden_states,
|
733 |
+
encoder_hidden_states=encoder_hidden_states,
|
734 |
+
attention_mask=attention_mask,
|
735 |
+
block_idx=block_idx,
|
736 |
+
additional_residuals=additional_residuals,
|
737 |
+
is_self_attn=is_self_attn,
|
738 |
+
**cross_attention_kwargs,
|
739 |
+
)
|
740 |
+
|
741 |
+
class ResidualXFormersAttnProcessor(XFormersAttnProcessor):
|
742 |
+
def __call__(
|
743 |
+
self,
|
744 |
+
attn: Attention,
|
745 |
+
hidden_states: torch.FloatTensor,
|
746 |
+
encoder_hidden_states: Optional[torch.FloatTensor] = None,
|
747 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
748 |
+
temb: Optional[torch.FloatTensor] = None,
|
749 |
+
block_idx: Optional[int] = None,
|
750 |
+
additional_residuals: Optional[Dict[str, torch.FloatTensor]] = None,
|
751 |
+
is_self_attn: Optional[bool] = None
|
752 |
+
):
|
753 |
+
residual = hidden_states
|
754 |
+
|
755 |
+
if attn.spatial_norm is not None:
|
756 |
+
hidden_states = attn.spatial_norm(hidden_states, temb)
|
757 |
+
|
758 |
+
input_ndim = hidden_states.ndim
|
759 |
+
|
760 |
+
if input_ndim == 4:
|
761 |
+
batch_size, channel, height, width = hidden_states.shape
|
762 |
+
hidden_states = hidden_states.view(batch_size, channel, height * width).transpose(1, 2)
|
763 |
+
|
764 |
+
batch_size, key_tokens, _ = (
|
765 |
+
hidden_states.shape if encoder_hidden_states is None else encoder_hidden_states.shape
|
766 |
+
)
|
767 |
+
|
768 |
+
attention_mask = attn.prepare_attention_mask(attention_mask, key_tokens, batch_size)
|
769 |
+
if attention_mask is not None:
|
770 |
+
# expand our mask's singleton query_tokens dimension:
|
771 |
+
# [batch*heads, 1, key_tokens] ->
|
772 |
+
# [batch*heads, query_tokens, key_tokens]
|
773 |
+
# so that it can be added as a bias onto the attention scores that xformers computes:
|
774 |
+
# [batch*heads, query_tokens, key_tokens]
|
775 |
+
# we do this explicitly because xformers doesn't broadcast the singleton dimension for us.
|
776 |
+
_, query_tokens, _ = hidden_states.shape
|
777 |
+
attention_mask = attention_mask.expand(-1, query_tokens, -1)
|
778 |
+
|
779 |
+
if attn.group_norm is not None:
|
780 |
+
hidden_states = attn.group_norm(hidden_states.transpose(1, 2)).transpose(1, 2)
|
781 |
+
|
782 |
+
query = attn.to_q(hidden_states)
|
783 |
+
|
784 |
+
# newly added
|
785 |
+
if is_self_attn and additional_residuals and f"block_{block_idx}_self_attn_q" in additional_residuals:
|
786 |
+
query = query + additional_residuals[f"block_{block_idx}_self_attn_q"]
|
787 |
+
elif not is_self_attn and additional_residuals and f"block_{block_idx}_cross_attn_q" in additional_residuals:
|
788 |
+
query = query + additional_residuals[f"block_{block_idx}_cross_attn_q"]
|
789 |
+
|
790 |
+
if encoder_hidden_states is None:
|
791 |
+
encoder_hidden_states = hidden_states
|
792 |
+
elif attn.norm_cross:
|
793 |
+
encoder_hidden_states = attn.norm_encoder_hidden_states(encoder_hidden_states)
|
794 |
+
|
795 |
+
if not is_self_attn and additional_residuals and f"block_{block_idx}_cross_attn_c" in additional_residuals:
|
796 |
+
not_uc = torch.abs(encoder_hidden_states - torch.zeros_like(encoder_hidden_states)).mean(dim=[1, 2], keepdim=True) < 1e-4
|
797 |
+
encoder_hidden_states = encoder_hidden_states + additional_residuals[f"block_{block_idx}_cross_attn_c"] * not_uc
|
798 |
+
# encoder_hidden_states[not_uc] = encoder_hidden_states[not_uc] + \
|
799 |
+
# additional_residuals[f"block_{block_idx}_cross_attn_c"][not_uc]
|
800 |
+
# encoder_hidden_states[~not_uc] = encoder_hidden_states[~not_uc] + \
|
801 |
+
# additional_residuals[f"block_{block_idx}_cross_attn_c"][~not_uc] * 0.
|
802 |
+
|
803 |
+
key = attn.to_k(encoder_hidden_states)
|
804 |
+
value = attn.to_v(encoder_hidden_states)
|
805 |
+
|
806 |
+
# newly added
|
807 |
+
if is_self_attn and additional_residuals and f"block_{block_idx}_self_attn_k" in additional_residuals:
|
808 |
+
key = key + additional_residuals[f"block_{block_idx}_self_attn_k"]
|
809 |
+
elif not is_self_attn and additional_residuals and f"block_{block_idx}_cross_attn_k" in additional_residuals:
|
810 |
+
key = key + additional_residuals[f"block_{block_idx}_cross_attn_k"]
|
811 |
+
|
812 |
+
if is_self_attn and additional_residuals and f"block_{block_idx}_self_attn_v" in additional_residuals:
|
813 |
+
value = value + additional_residuals[f"block_{block_idx}_self_attn_v"]
|
814 |
+
elif not is_self_attn and additional_residuals and f"block_{block_idx}_cross_attn_v" in additional_residuals:
|
815 |
+
value = value + additional_residuals[f"block_{block_idx}_cross_attn_v"]
|
816 |
+
|
817 |
+
query = attn.head_to_batch_dim(query).contiguous()
|
818 |
+
key = attn.head_to_batch_dim(key).contiguous()
|
819 |
+
value = attn.head_to_batch_dim(value).contiguous()
|
820 |
+
|
821 |
+
hidden_states = xformers.ops.memory_efficient_attention(
|
822 |
+
query, key, value, attn_bias=attention_mask, op=self.attention_op, scale=attn.scale
|
823 |
+
)
|
824 |
+
hidden_states = hidden_states.to(query.dtype)
|
825 |
+
hidden_states = attn.batch_to_head_dim(hidden_states)
|
826 |
+
|
827 |
+
# linear proj
|
828 |
+
hidden_states = attn.to_out[0](hidden_states)
|
829 |
+
# dropout
|
830 |
+
hidden_states = attn.to_out[1](hidden_states)
|
831 |
+
|
832 |
+
if input_ndim == 4:
|
833 |
+
hidden_states = hidden_states.transpose(-1, -2).reshape(batch_size, channel, height, width)
|
834 |
+
|
835 |
+
if attn.residual_connection:
|
836 |
+
hidden_states = hidden_states + residual
|
837 |
+
|
838 |
+
hidden_states = hidden_states / attn.rescale_output_factor
|
839 |
+
|
840 |
+
return hidden_states
|
src/models/motion_module.py
ADDED
@@ -0,0 +1,388 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapt from https://github.com/guoyww/AnimateDiff/blob/main/animatediff/models/motion_module.py
|
2 |
+
import math
|
3 |
+
from dataclasses import dataclass
|
4 |
+
from typing import Callable, Optional
|
5 |
+
|
6 |
+
import torch
|
7 |
+
from diffusers.models.attention import FeedForward
|
8 |
+
from diffusers.models.attention_processor import Attention, AttnProcessor
|
9 |
+
from diffusers.utils import BaseOutput
|
10 |
+
from diffusers.utils.import_utils import is_xformers_available
|
11 |
+
from einops import rearrange, repeat
|
12 |
+
from torch import nn
|
13 |
+
|
14 |
+
|
15 |
+
def zero_module(module):
|
16 |
+
# Zero out the parameters of a module and return it.
|
17 |
+
for p in module.parameters():
|
18 |
+
p.detach().zero_()
|
19 |
+
return module
|
20 |
+
|
21 |
+
|
22 |
+
@dataclass
|
23 |
+
class TemporalTransformer3DModelOutput(BaseOutput):
|
24 |
+
sample: torch.FloatTensor
|
25 |
+
|
26 |
+
|
27 |
+
if is_xformers_available():
|
28 |
+
import xformers
|
29 |
+
import xformers.ops
|
30 |
+
else:
|
31 |
+
xformers = None
|
32 |
+
|
33 |
+
|
34 |
+
def get_motion_module(in_channels, motion_module_type: str, motion_module_kwargs: dict):
|
35 |
+
if motion_module_type == "Vanilla":
|
36 |
+
return VanillaTemporalModule(
|
37 |
+
in_channels=in_channels,
|
38 |
+
**motion_module_kwargs,
|
39 |
+
)
|
40 |
+
else:
|
41 |
+
raise ValueError
|
42 |
+
|
43 |
+
|
44 |
+
class VanillaTemporalModule(nn.Module):
|
45 |
+
def __init__(
|
46 |
+
self,
|
47 |
+
in_channels,
|
48 |
+
num_attention_heads=8,
|
49 |
+
num_transformer_block=2,
|
50 |
+
attention_block_types=("Temporal_Self", "Temporal_Self"),
|
51 |
+
cross_frame_attention_mode=None,
|
52 |
+
temporal_position_encoding=False,
|
53 |
+
temporal_position_encoding_max_len=24,
|
54 |
+
temporal_attention_dim_div=1,
|
55 |
+
zero_initialize=True,
|
56 |
+
):
|
57 |
+
super().__init__()
|
58 |
+
|
59 |
+
self.temporal_transformer = TemporalTransformer3DModel(
|
60 |
+
in_channels=in_channels,
|
61 |
+
num_attention_heads=num_attention_heads,
|
62 |
+
attention_head_dim=in_channels
|
63 |
+
// num_attention_heads
|
64 |
+
// temporal_attention_dim_div,
|
65 |
+
num_layers=num_transformer_block,
|
66 |
+
attention_block_types=attention_block_types,
|
67 |
+
cross_frame_attention_mode=cross_frame_attention_mode,
|
68 |
+
temporal_position_encoding=temporal_position_encoding,
|
69 |
+
temporal_position_encoding_max_len=temporal_position_encoding_max_len,
|
70 |
+
)
|
71 |
+
|
72 |
+
if zero_initialize:
|
73 |
+
self.temporal_transformer.proj_out = zero_module(
|
74 |
+
self.temporal_transformer.proj_out
|
75 |
+
)
|
76 |
+
|
77 |
+
def forward(
|
78 |
+
self,
|
79 |
+
input_tensor,
|
80 |
+
temb,
|
81 |
+
encoder_hidden_states,
|
82 |
+
attention_mask=None,
|
83 |
+
anchor_frame_idx=None,
|
84 |
+
):
|
85 |
+
hidden_states = input_tensor
|
86 |
+
hidden_states = self.temporal_transformer(
|
87 |
+
hidden_states, encoder_hidden_states, attention_mask
|
88 |
+
)
|
89 |
+
|
90 |
+
output = hidden_states
|
91 |
+
return output
|
92 |
+
|
93 |
+
|
94 |
+
class TemporalTransformer3DModel(nn.Module):
|
95 |
+
def __init__(
|
96 |
+
self,
|
97 |
+
in_channels,
|
98 |
+
num_attention_heads,
|
99 |
+
attention_head_dim,
|
100 |
+
num_layers,
|
101 |
+
attention_block_types=(
|
102 |
+
"Temporal_Self",
|
103 |
+
"Temporal_Self",
|
104 |
+
),
|
105 |
+
dropout=0.0,
|
106 |
+
norm_num_groups=32,
|
107 |
+
cross_attention_dim=768,
|
108 |
+
activation_fn="geglu",
|
109 |
+
attention_bias=False,
|
110 |
+
upcast_attention=False,
|
111 |
+
cross_frame_attention_mode=None,
|
112 |
+
temporal_position_encoding=False,
|
113 |
+
temporal_position_encoding_max_len=24,
|
114 |
+
):
|
115 |
+
super().__init__()
|
116 |
+
|
117 |
+
inner_dim = num_attention_heads * attention_head_dim
|
118 |
+
|
119 |
+
self.norm = torch.nn.GroupNorm(
|
120 |
+
num_groups=norm_num_groups, num_channels=in_channels, eps=1e-6, affine=True
|
121 |
+
)
|
122 |
+
self.proj_in = nn.Linear(in_channels, inner_dim)
|
123 |
+
|
124 |
+
self.transformer_blocks = nn.ModuleList(
|
125 |
+
[
|
126 |
+
TemporalTransformerBlock(
|
127 |
+
dim=inner_dim,
|
128 |
+
num_attention_heads=num_attention_heads,
|
129 |
+
attention_head_dim=attention_head_dim,
|
130 |
+
attention_block_types=attention_block_types,
|
131 |
+
dropout=dropout,
|
132 |
+
norm_num_groups=norm_num_groups,
|
133 |
+
cross_attention_dim=cross_attention_dim,
|
134 |
+
activation_fn=activation_fn,
|
135 |
+
attention_bias=attention_bias,
|
136 |
+
upcast_attention=upcast_attention,
|
137 |
+
cross_frame_attention_mode=cross_frame_attention_mode,
|
138 |
+
temporal_position_encoding=temporal_position_encoding,
|
139 |
+
temporal_position_encoding_max_len=temporal_position_encoding_max_len,
|
140 |
+
)
|
141 |
+
for d in range(num_layers)
|
142 |
+
]
|
143 |
+
)
|
144 |
+
self.proj_out = nn.Linear(inner_dim, in_channels)
|
145 |
+
|
146 |
+
def forward(self, hidden_states, encoder_hidden_states=None, attention_mask=None):
|
147 |
+
assert (
|
148 |
+
hidden_states.dim() == 5
|
149 |
+
), f"Expected hidden_states to have ndim=5, but got ndim={hidden_states.dim()}."
|
150 |
+
video_length = hidden_states.shape[2]
|
151 |
+
hidden_states = rearrange(hidden_states, "b c f h w -> (b f) c h w")
|
152 |
+
|
153 |
+
batch, channel, height, weight = hidden_states.shape
|
154 |
+
residual = hidden_states
|
155 |
+
|
156 |
+
hidden_states = self.norm(hidden_states)
|
157 |
+
inner_dim = hidden_states.shape[1]
|
158 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(
|
159 |
+
batch, height * weight, inner_dim
|
160 |
+
)
|
161 |
+
hidden_states = self.proj_in(hidden_states)
|
162 |
+
|
163 |
+
# Transformer Blocks
|
164 |
+
for block in self.transformer_blocks:
|
165 |
+
hidden_states = block(
|
166 |
+
hidden_states,
|
167 |
+
encoder_hidden_states=encoder_hidden_states,
|
168 |
+
video_length=video_length,
|
169 |
+
)
|
170 |
+
|
171 |
+
# output
|
172 |
+
hidden_states = self.proj_out(hidden_states)
|
173 |
+
hidden_states = (
|
174 |
+
hidden_states.reshape(batch, height, weight, inner_dim)
|
175 |
+
.permute(0, 3, 1, 2)
|
176 |
+
.contiguous()
|
177 |
+
)
|
178 |
+
|
179 |
+
output = hidden_states + residual
|
180 |
+
output = rearrange(output, "(b f) c h w -> b c f h w", f=video_length)
|
181 |
+
|
182 |
+
return output
|
183 |
+
|
184 |
+
|
185 |
+
class TemporalTransformerBlock(nn.Module):
|
186 |
+
def __init__(
|
187 |
+
self,
|
188 |
+
dim,
|
189 |
+
num_attention_heads,
|
190 |
+
attention_head_dim,
|
191 |
+
attention_block_types=(
|
192 |
+
"Temporal_Self",
|
193 |
+
"Temporal_Self",
|
194 |
+
),
|
195 |
+
dropout=0.0,
|
196 |
+
norm_num_groups=32,
|
197 |
+
cross_attention_dim=768,
|
198 |
+
activation_fn="geglu",
|
199 |
+
attention_bias=False,
|
200 |
+
upcast_attention=False,
|
201 |
+
cross_frame_attention_mode=None,
|
202 |
+
temporal_position_encoding=False,
|
203 |
+
temporal_position_encoding_max_len=24,
|
204 |
+
):
|
205 |
+
super().__init__()
|
206 |
+
|
207 |
+
attention_blocks = []
|
208 |
+
norms = []
|
209 |
+
|
210 |
+
for block_name in attention_block_types:
|
211 |
+
attention_blocks.append(
|
212 |
+
VersatileAttention(
|
213 |
+
attention_mode=block_name.split("_")[0],
|
214 |
+
cross_attention_dim=cross_attention_dim
|
215 |
+
if block_name.endswith("_Cross")
|
216 |
+
else None,
|
217 |
+
query_dim=dim,
|
218 |
+
heads=num_attention_heads,
|
219 |
+
dim_head=attention_head_dim,
|
220 |
+
dropout=dropout,
|
221 |
+
bias=attention_bias,
|
222 |
+
upcast_attention=upcast_attention,
|
223 |
+
cross_frame_attention_mode=cross_frame_attention_mode,
|
224 |
+
temporal_position_encoding=temporal_position_encoding,
|
225 |
+
temporal_position_encoding_max_len=temporal_position_encoding_max_len,
|
226 |
+
)
|
227 |
+
)
|
228 |
+
norms.append(nn.LayerNorm(dim))
|
229 |
+
|
230 |
+
self.attention_blocks = nn.ModuleList(attention_blocks)
|
231 |
+
self.norms = nn.ModuleList(norms)
|
232 |
+
|
233 |
+
self.ff = FeedForward(dim, dropout=dropout, activation_fn=activation_fn)
|
234 |
+
self.ff_norm = nn.LayerNorm(dim)
|
235 |
+
|
236 |
+
def forward(
|
237 |
+
self,
|
238 |
+
hidden_states,
|
239 |
+
encoder_hidden_states=None,
|
240 |
+
attention_mask=None,
|
241 |
+
video_length=None,
|
242 |
+
):
|
243 |
+
for attention_block, norm in zip(self.attention_blocks, self.norms):
|
244 |
+
norm_hidden_states = norm(hidden_states)
|
245 |
+
hidden_states = (
|
246 |
+
attention_block(
|
247 |
+
norm_hidden_states,
|
248 |
+
encoder_hidden_states=encoder_hidden_states
|
249 |
+
if attention_block.is_cross_attention
|
250 |
+
else None,
|
251 |
+
video_length=video_length,
|
252 |
+
)
|
253 |
+
+ hidden_states
|
254 |
+
)
|
255 |
+
|
256 |
+
hidden_states = self.ff(self.ff_norm(hidden_states)) + hidden_states
|
257 |
+
|
258 |
+
output = hidden_states
|
259 |
+
return output
|
260 |
+
|
261 |
+
|
262 |
+
class PositionalEncoding(nn.Module):
|
263 |
+
def __init__(self, d_model, dropout=0.0, max_len=24):
|
264 |
+
super().__init__()
|
265 |
+
self.dropout = nn.Dropout(p=dropout)
|
266 |
+
position = torch.arange(max_len).unsqueeze(1)
|
267 |
+
div_term = torch.exp(
|
268 |
+
torch.arange(0, d_model, 2) * (-math.log(10000.0) / d_model)
|
269 |
+
)
|
270 |
+
pe = torch.zeros(1, max_len, d_model)
|
271 |
+
pe[0, :, 0::2] = torch.sin(position * div_term)
|
272 |
+
pe[0, :, 1::2] = torch.cos(position * div_term)
|
273 |
+
self.register_buffer("pe", pe)
|
274 |
+
|
275 |
+
def forward(self, x):
|
276 |
+
x = x + self.pe[:, : x.size(1)]
|
277 |
+
return self.dropout(x)
|
278 |
+
|
279 |
+
|
280 |
+
class VersatileAttention(Attention):
|
281 |
+
def __init__(
|
282 |
+
self,
|
283 |
+
attention_mode=None,
|
284 |
+
cross_frame_attention_mode=None,
|
285 |
+
temporal_position_encoding=False,
|
286 |
+
temporal_position_encoding_max_len=24,
|
287 |
+
*args,
|
288 |
+
**kwargs,
|
289 |
+
):
|
290 |
+
super().__init__(*args, **kwargs)
|
291 |
+
assert attention_mode == "Temporal"
|
292 |
+
|
293 |
+
self.attention_mode = attention_mode
|
294 |
+
self.is_cross_attention = kwargs["cross_attention_dim"] is not None
|
295 |
+
|
296 |
+
self.pos_encoder = (
|
297 |
+
PositionalEncoding(
|
298 |
+
kwargs["query_dim"],
|
299 |
+
dropout=0.0,
|
300 |
+
max_len=temporal_position_encoding_max_len,
|
301 |
+
)
|
302 |
+
if (temporal_position_encoding and attention_mode == "Temporal")
|
303 |
+
else None
|
304 |
+
)
|
305 |
+
|
306 |
+
def extra_repr(self):
|
307 |
+
return f"(Module Info) Attention_Mode: {self.attention_mode}, Is_Cross_Attention: {self.is_cross_attention}"
|
308 |
+
|
309 |
+
def set_use_memory_efficient_attention_xformers(
|
310 |
+
self,
|
311 |
+
use_memory_efficient_attention_xformers: bool,
|
312 |
+
attention_op: Optional[Callable] = None,
|
313 |
+
):
|
314 |
+
if use_memory_efficient_attention_xformers:
|
315 |
+
if not is_xformers_available():
|
316 |
+
raise ModuleNotFoundError(
|
317 |
+
(
|
318 |
+
"Refer to https://github.com/facebookresearch/xformers for more information on how to install"
|
319 |
+
" xformers"
|
320 |
+
),
|
321 |
+
name="xformers",
|
322 |
+
)
|
323 |
+
elif not torch.cuda.is_available():
|
324 |
+
raise ValueError(
|
325 |
+
"torch.cuda.is_available() should be True but is False. xformers' memory efficient attention is"
|
326 |
+
" only available for GPU "
|
327 |
+
)
|
328 |
+
else:
|
329 |
+
try:
|
330 |
+
# Make sure we can run the memory efficient attention
|
331 |
+
_ = xformers.ops.memory_efficient_attention(
|
332 |
+
torch.randn((1, 2, 40), device="cuda"),
|
333 |
+
torch.randn((1, 2, 40), device="cuda"),
|
334 |
+
torch.randn((1, 2, 40), device="cuda"),
|
335 |
+
)
|
336 |
+
except Exception as e:
|
337 |
+
raise e
|
338 |
+
|
339 |
+
# XFormersAttnProcessor corrupts video generation and work with Pytorch 1.13.
|
340 |
+
# Pytorch 2.0.1 AttnProcessor works the same as XFormersAttnProcessor in Pytorch 1.13.
|
341 |
+
# You don't need XFormersAttnProcessor here.
|
342 |
+
# processor = XFormersAttnProcessor(
|
343 |
+
# attention_op=attention_op,
|
344 |
+
# )
|
345 |
+
processor = AttnProcessor()
|
346 |
+
else:
|
347 |
+
processor = AttnProcessor()
|
348 |
+
|
349 |
+
self.set_processor(processor)
|
350 |
+
|
351 |
+
def forward(
|
352 |
+
self,
|
353 |
+
hidden_states,
|
354 |
+
encoder_hidden_states=None,
|
355 |
+
attention_mask=None,
|
356 |
+
video_length=None,
|
357 |
+
**cross_attention_kwargs,
|
358 |
+
):
|
359 |
+
if self.attention_mode == "Temporal":
|
360 |
+
d = hidden_states.shape[1] # d means HxW
|
361 |
+
hidden_states = rearrange(
|
362 |
+
hidden_states, "(b f) d c -> (b d) f c", f=video_length
|
363 |
+
)
|
364 |
+
|
365 |
+
if self.pos_encoder is not None:
|
366 |
+
hidden_states = self.pos_encoder(hidden_states)
|
367 |
+
|
368 |
+
encoder_hidden_states = (
|
369 |
+
repeat(encoder_hidden_states, "b n c -> (b d) n c", d=d)
|
370 |
+
if encoder_hidden_states is not None
|
371 |
+
else encoder_hidden_states
|
372 |
+
)
|
373 |
+
|
374 |
+
else:
|
375 |
+
raise NotImplementedError
|
376 |
+
|
377 |
+
hidden_states = self.processor(
|
378 |
+
self,
|
379 |
+
hidden_states,
|
380 |
+
encoder_hidden_states=encoder_hidden_states,
|
381 |
+
attention_mask=attention_mask,
|
382 |
+
**cross_attention_kwargs,
|
383 |
+
)
|
384 |
+
|
385 |
+
if self.attention_mode == "Temporal":
|
386 |
+
hidden_states = rearrange(hidden_states, "(b d) f c -> (b f) d c", d=d)
|
387 |
+
|
388 |
+
return hidden_states
|
src/models/mutual_self_attention.py
ADDED
@@ -0,0 +1,363 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/magic-research/magic-animate/blob/main/magicanimate/models/mutual_self_attention.py
|
2 |
+
from typing import Any, Dict, Optional
|
3 |
+
|
4 |
+
import torch
|
5 |
+
from einops import rearrange
|
6 |
+
|
7 |
+
from src.models.attention import TemporalBasicTransformerBlock
|
8 |
+
|
9 |
+
from .attention import BasicTransformerBlock
|
10 |
+
|
11 |
+
|
12 |
+
def torch_dfs(model: torch.nn.Module):
|
13 |
+
result = [model]
|
14 |
+
for child in model.children():
|
15 |
+
result += torch_dfs(child)
|
16 |
+
return result
|
17 |
+
|
18 |
+
|
19 |
+
class ReferenceAttentionControl:
|
20 |
+
def __init__(
|
21 |
+
self,
|
22 |
+
unet,
|
23 |
+
mode="write",
|
24 |
+
do_classifier_free_guidance=False,
|
25 |
+
attention_auto_machine_weight=float("inf"),
|
26 |
+
gn_auto_machine_weight=1.0,
|
27 |
+
style_fidelity=1.0,
|
28 |
+
reference_attn=True,
|
29 |
+
reference_adain=False,
|
30 |
+
fusion_blocks="midup",
|
31 |
+
batch_size=1,
|
32 |
+
) -> None:
|
33 |
+
# 10. Modify self attention and group norm
|
34 |
+
self.unet = unet
|
35 |
+
assert mode in ["read", "write"]
|
36 |
+
assert fusion_blocks in ["midup", "full"]
|
37 |
+
self.reference_attn = reference_attn
|
38 |
+
self.reference_adain = reference_adain
|
39 |
+
self.fusion_blocks = fusion_blocks
|
40 |
+
self.register_reference_hooks(
|
41 |
+
mode,
|
42 |
+
do_classifier_free_guidance,
|
43 |
+
attention_auto_machine_weight,
|
44 |
+
gn_auto_machine_weight,
|
45 |
+
style_fidelity,
|
46 |
+
reference_attn,
|
47 |
+
reference_adain,
|
48 |
+
fusion_blocks,
|
49 |
+
batch_size=batch_size,
|
50 |
+
)
|
51 |
+
|
52 |
+
def register_reference_hooks(
|
53 |
+
self,
|
54 |
+
mode,
|
55 |
+
do_classifier_free_guidance,
|
56 |
+
attention_auto_machine_weight,
|
57 |
+
gn_auto_machine_weight,
|
58 |
+
style_fidelity,
|
59 |
+
reference_attn,
|
60 |
+
reference_adain,
|
61 |
+
dtype=torch.float16,
|
62 |
+
batch_size=1,
|
63 |
+
num_images_per_prompt=1,
|
64 |
+
device=torch.device("cpu"),
|
65 |
+
fusion_blocks="midup",
|
66 |
+
):
|
67 |
+
MODE = mode
|
68 |
+
do_classifier_free_guidance = do_classifier_free_guidance
|
69 |
+
attention_auto_machine_weight = attention_auto_machine_weight
|
70 |
+
gn_auto_machine_weight = gn_auto_machine_weight
|
71 |
+
style_fidelity = style_fidelity
|
72 |
+
reference_attn = reference_attn
|
73 |
+
reference_adain = reference_adain
|
74 |
+
fusion_blocks = fusion_blocks
|
75 |
+
num_images_per_prompt = num_images_per_prompt
|
76 |
+
dtype = dtype
|
77 |
+
if do_classifier_free_guidance:
|
78 |
+
uc_mask = (
|
79 |
+
torch.Tensor(
|
80 |
+
[1] * batch_size * num_images_per_prompt * 16
|
81 |
+
+ [0] * batch_size * num_images_per_prompt * 16
|
82 |
+
)
|
83 |
+
.to(device)
|
84 |
+
.bool()
|
85 |
+
)
|
86 |
+
else:
|
87 |
+
uc_mask = (
|
88 |
+
torch.Tensor([0] * batch_size * num_images_per_prompt * 2)
|
89 |
+
.to(device)
|
90 |
+
.bool()
|
91 |
+
)
|
92 |
+
|
93 |
+
def hacked_basic_transformer_inner_forward(
|
94 |
+
self,
|
95 |
+
hidden_states: torch.FloatTensor,
|
96 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
97 |
+
encoder_hidden_states: Optional[torch.FloatTensor] = None,
|
98 |
+
encoder_attention_mask: Optional[torch.FloatTensor] = None,
|
99 |
+
timestep: Optional[torch.LongTensor] = None,
|
100 |
+
cross_attention_kwargs: Dict[str, Any] = None,
|
101 |
+
class_labels: Optional[torch.LongTensor] = None,
|
102 |
+
video_length=None,
|
103 |
+
):
|
104 |
+
if self.use_ada_layer_norm: # False
|
105 |
+
norm_hidden_states = self.norm1(hidden_states, timestep)
|
106 |
+
elif self.use_ada_layer_norm_zero:
|
107 |
+
(
|
108 |
+
norm_hidden_states,
|
109 |
+
gate_msa,
|
110 |
+
shift_mlp,
|
111 |
+
scale_mlp,
|
112 |
+
gate_mlp,
|
113 |
+
) = self.norm1(
|
114 |
+
hidden_states,
|
115 |
+
timestep,
|
116 |
+
class_labels,
|
117 |
+
hidden_dtype=hidden_states.dtype,
|
118 |
+
)
|
119 |
+
else:
|
120 |
+
norm_hidden_states = self.norm1(hidden_states)
|
121 |
+
|
122 |
+
# 1. Self-Attention
|
123 |
+
# self.only_cross_attention = False
|
124 |
+
cross_attention_kwargs = (
|
125 |
+
cross_attention_kwargs if cross_attention_kwargs is not None else {}
|
126 |
+
)
|
127 |
+
if self.only_cross_attention:
|
128 |
+
attn_output = self.attn1(
|
129 |
+
norm_hidden_states,
|
130 |
+
encoder_hidden_states=encoder_hidden_states
|
131 |
+
if self.only_cross_attention
|
132 |
+
else None,
|
133 |
+
attention_mask=attention_mask,
|
134 |
+
**cross_attention_kwargs,
|
135 |
+
)
|
136 |
+
else:
|
137 |
+
if MODE == "write":
|
138 |
+
self.bank.append(norm_hidden_states.clone())
|
139 |
+
attn_output = self.attn1(
|
140 |
+
norm_hidden_states,
|
141 |
+
encoder_hidden_states=encoder_hidden_states
|
142 |
+
if self.only_cross_attention
|
143 |
+
else None,
|
144 |
+
attention_mask=attention_mask,
|
145 |
+
**cross_attention_kwargs,
|
146 |
+
)
|
147 |
+
if MODE == "read":
|
148 |
+
bank_fea = [
|
149 |
+
rearrange(
|
150 |
+
d.unsqueeze(1).repeat(1, video_length, 1, 1),
|
151 |
+
"b t l c -> (b t) l c",
|
152 |
+
)
|
153 |
+
for d in self.bank
|
154 |
+
]
|
155 |
+
modify_norm_hidden_states = torch.cat(
|
156 |
+
[norm_hidden_states] + bank_fea, dim=1
|
157 |
+
)
|
158 |
+
hidden_states_uc = (
|
159 |
+
self.attn1(
|
160 |
+
norm_hidden_states,
|
161 |
+
encoder_hidden_states=modify_norm_hidden_states,
|
162 |
+
attention_mask=attention_mask,
|
163 |
+
)
|
164 |
+
+ hidden_states
|
165 |
+
)
|
166 |
+
if do_classifier_free_guidance:
|
167 |
+
hidden_states_c = hidden_states_uc.clone()
|
168 |
+
_uc_mask = uc_mask.clone()
|
169 |
+
if hidden_states.shape[0] != _uc_mask.shape[0]:
|
170 |
+
_uc_mask = (
|
171 |
+
torch.Tensor(
|
172 |
+
[1] * (hidden_states.shape[0] // 2)
|
173 |
+
+ [0] * (hidden_states.shape[0] // 2)
|
174 |
+
)
|
175 |
+
.to(device)
|
176 |
+
.bool()
|
177 |
+
)
|
178 |
+
hidden_states_c[_uc_mask] = (
|
179 |
+
self.attn1(
|
180 |
+
norm_hidden_states[_uc_mask],
|
181 |
+
encoder_hidden_states=norm_hidden_states[_uc_mask],
|
182 |
+
attention_mask=attention_mask,
|
183 |
+
)
|
184 |
+
+ hidden_states[_uc_mask]
|
185 |
+
)
|
186 |
+
hidden_states = hidden_states_c.clone()
|
187 |
+
else:
|
188 |
+
hidden_states = hidden_states_uc
|
189 |
+
|
190 |
+
# self.bank.clear()
|
191 |
+
if self.attn2 is not None:
|
192 |
+
# Cross-Attention
|
193 |
+
norm_hidden_states = (
|
194 |
+
self.norm2(hidden_states, timestep)
|
195 |
+
if self.use_ada_layer_norm
|
196 |
+
else self.norm2(hidden_states)
|
197 |
+
)
|
198 |
+
hidden_states = (
|
199 |
+
self.attn2(
|
200 |
+
norm_hidden_states,
|
201 |
+
encoder_hidden_states=encoder_hidden_states,
|
202 |
+
attention_mask=attention_mask,
|
203 |
+
)
|
204 |
+
+ hidden_states
|
205 |
+
)
|
206 |
+
|
207 |
+
# Feed-forward
|
208 |
+
hidden_states = self.ff(self.norm3(hidden_states)) + hidden_states
|
209 |
+
|
210 |
+
# Temporal-Attention
|
211 |
+
if self.unet_use_temporal_attention:
|
212 |
+
d = hidden_states.shape[1]
|
213 |
+
hidden_states = rearrange(
|
214 |
+
hidden_states, "(b f) d c -> (b d) f c", f=video_length
|
215 |
+
)
|
216 |
+
norm_hidden_states = (
|
217 |
+
self.norm_temp(hidden_states, timestep)
|
218 |
+
if self.use_ada_layer_norm
|
219 |
+
else self.norm_temp(hidden_states)
|
220 |
+
)
|
221 |
+
hidden_states = (
|
222 |
+
self.attn_temp(norm_hidden_states) + hidden_states
|
223 |
+
)
|
224 |
+
hidden_states = rearrange(
|
225 |
+
hidden_states, "(b d) f c -> (b f) d c", d=d
|
226 |
+
)
|
227 |
+
|
228 |
+
return hidden_states
|
229 |
+
|
230 |
+
if self.use_ada_layer_norm_zero:
|
231 |
+
attn_output = gate_msa.unsqueeze(1) * attn_output
|
232 |
+
hidden_states = attn_output + hidden_states
|
233 |
+
|
234 |
+
if self.attn2 is not None:
|
235 |
+
norm_hidden_states = (
|
236 |
+
self.norm2(hidden_states, timestep)
|
237 |
+
if self.use_ada_layer_norm
|
238 |
+
else self.norm2(hidden_states)
|
239 |
+
)
|
240 |
+
|
241 |
+
# 2. Cross-Attention
|
242 |
+
attn_output = self.attn2(
|
243 |
+
norm_hidden_states,
|
244 |
+
encoder_hidden_states=encoder_hidden_states,
|
245 |
+
attention_mask=encoder_attention_mask,
|
246 |
+
**cross_attention_kwargs,
|
247 |
+
)
|
248 |
+
hidden_states = attn_output + hidden_states
|
249 |
+
|
250 |
+
# 3. Feed-forward
|
251 |
+
norm_hidden_states = self.norm3(hidden_states)
|
252 |
+
|
253 |
+
if self.use_ada_layer_norm_zero:
|
254 |
+
norm_hidden_states = (
|
255 |
+
norm_hidden_states * (1 + scale_mlp[:, None]) + shift_mlp[:, None]
|
256 |
+
)
|
257 |
+
|
258 |
+
ff_output = self.ff(norm_hidden_states)
|
259 |
+
|
260 |
+
if self.use_ada_layer_norm_zero:
|
261 |
+
ff_output = gate_mlp.unsqueeze(1) * ff_output
|
262 |
+
|
263 |
+
hidden_states = ff_output + hidden_states
|
264 |
+
|
265 |
+
return hidden_states
|
266 |
+
|
267 |
+
if self.reference_attn:
|
268 |
+
if self.fusion_blocks == "midup":
|
269 |
+
attn_modules = [
|
270 |
+
module
|
271 |
+
for module in (
|
272 |
+
torch_dfs(self.unet.mid_block) + torch_dfs(self.unet.up_blocks)
|
273 |
+
)
|
274 |
+
if isinstance(module, BasicTransformerBlock)
|
275 |
+
or isinstance(module, TemporalBasicTransformerBlock)
|
276 |
+
]
|
277 |
+
elif self.fusion_blocks == "full":
|
278 |
+
attn_modules = [
|
279 |
+
module
|
280 |
+
for module in torch_dfs(self.unet)
|
281 |
+
if isinstance(module, BasicTransformerBlock)
|
282 |
+
or isinstance(module, TemporalBasicTransformerBlock)
|
283 |
+
]
|
284 |
+
attn_modules = sorted(
|
285 |
+
attn_modules, key=lambda x: -x.norm1.normalized_shape[0]
|
286 |
+
)
|
287 |
+
|
288 |
+
for i, module in enumerate(attn_modules):
|
289 |
+
module._original_inner_forward = module.forward
|
290 |
+
if isinstance(module, BasicTransformerBlock):
|
291 |
+
module.forward = hacked_basic_transformer_inner_forward.__get__(
|
292 |
+
module, BasicTransformerBlock
|
293 |
+
)
|
294 |
+
if isinstance(module, TemporalBasicTransformerBlock):
|
295 |
+
module.forward = hacked_basic_transformer_inner_forward.__get__(
|
296 |
+
module, TemporalBasicTransformerBlock
|
297 |
+
)
|
298 |
+
|
299 |
+
module.bank = []
|
300 |
+
module.attn_weight = float(i) / float(len(attn_modules))
|
301 |
+
|
302 |
+
def update(self, writer, dtype=torch.float16):
|
303 |
+
if self.reference_attn:
|
304 |
+
if self.fusion_blocks == "midup":
|
305 |
+
reader_attn_modules = [
|
306 |
+
module
|
307 |
+
for module in (
|
308 |
+
torch_dfs(self.unet.mid_block) + torch_dfs(self.unet.up_blocks)
|
309 |
+
)
|
310 |
+
if isinstance(module, TemporalBasicTransformerBlock)
|
311 |
+
]
|
312 |
+
writer_attn_modules = [
|
313 |
+
module
|
314 |
+
for module in (
|
315 |
+
torch_dfs(writer.unet.mid_block)
|
316 |
+
+ torch_dfs(writer.unet.up_blocks)
|
317 |
+
)
|
318 |
+
if isinstance(module, BasicTransformerBlock)
|
319 |
+
]
|
320 |
+
elif self.fusion_blocks == "full":
|
321 |
+
reader_attn_modules = [
|
322 |
+
module
|
323 |
+
for module in torch_dfs(self.unet)
|
324 |
+
if isinstance(module, TemporalBasicTransformerBlock)
|
325 |
+
]
|
326 |
+
writer_attn_modules = [
|
327 |
+
module
|
328 |
+
for module in torch_dfs(writer.unet)
|
329 |
+
if isinstance(module, BasicTransformerBlock)
|
330 |
+
]
|
331 |
+
reader_attn_modules = sorted(
|
332 |
+
reader_attn_modules, key=lambda x: -x.norm1.normalized_shape[0]
|
333 |
+
)
|
334 |
+
writer_attn_modules = sorted(
|
335 |
+
writer_attn_modules, key=lambda x: -x.norm1.normalized_shape[0]
|
336 |
+
)
|
337 |
+
for r, w in zip(reader_attn_modules, writer_attn_modules):
|
338 |
+
r.bank = [v.clone().to(dtype) for v in w.bank]
|
339 |
+
# w.bank.clear()
|
340 |
+
|
341 |
+
def clear(self):
|
342 |
+
if self.reference_attn:
|
343 |
+
if self.fusion_blocks == "midup":
|
344 |
+
reader_attn_modules = [
|
345 |
+
module
|
346 |
+
for module in (
|
347 |
+
torch_dfs(self.unet.mid_block) + torch_dfs(self.unet.up_blocks)
|
348 |
+
)
|
349 |
+
if isinstance(module, BasicTransformerBlock)
|
350 |
+
or isinstance(module, TemporalBasicTransformerBlock)
|
351 |
+
]
|
352 |
+
elif self.fusion_blocks == "full":
|
353 |
+
reader_attn_modules = [
|
354 |
+
module
|
355 |
+
for module in torch_dfs(self.unet)
|
356 |
+
if isinstance(module, BasicTransformerBlock)
|
357 |
+
or isinstance(module, TemporalBasicTransformerBlock)
|
358 |
+
]
|
359 |
+
reader_attn_modules = sorted(
|
360 |
+
reader_attn_modules, key=lambda x: -x.norm1.normalized_shape[0]
|
361 |
+
)
|
362 |
+
for r in reader_attn_modules:
|
363 |
+
r.bank.clear()
|
src/models/pose_guider.py
ADDED
@@ -0,0 +1,329 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import torch
|
3 |
+
import torch.nn as nn
|
4 |
+
import torch.nn.init as init
|
5 |
+
from einops import rearrange
|
6 |
+
import numpy as np
|
7 |
+
from diffusers.models.modeling_utils import ModelMixin
|
8 |
+
|
9 |
+
from typing import Any, Dict, Optional
|
10 |
+
from src.models.attention import BasicTransformerBlock
|
11 |
+
|
12 |
+
|
13 |
+
class PoseGuider(ModelMixin):
|
14 |
+
def __init__(self, noise_latent_channels=320, use_ca=True):
|
15 |
+
super(PoseGuider, self).__init__()
|
16 |
+
|
17 |
+
self.use_ca = use_ca
|
18 |
+
|
19 |
+
self.conv_layers = nn.Sequential(
|
20 |
+
nn.Conv2d(in_channels=3, out_channels=3, kernel_size=3, padding=1),
|
21 |
+
nn.BatchNorm2d(3),
|
22 |
+
nn.ReLU(),
|
23 |
+
nn.Conv2d(in_channels=3, out_channels=16, kernel_size=4, stride=2, padding=1),
|
24 |
+
nn.BatchNorm2d(16),
|
25 |
+
nn.ReLU(),
|
26 |
+
|
27 |
+
nn.Conv2d(in_channels=16, out_channels=16, kernel_size=3, padding=1),
|
28 |
+
nn.BatchNorm2d(16),
|
29 |
+
nn.ReLU(),
|
30 |
+
nn.Conv2d(in_channels=16, out_channels=32, kernel_size=4, stride=2, padding=1),
|
31 |
+
nn.BatchNorm2d(32),
|
32 |
+
nn.ReLU(),
|
33 |
+
|
34 |
+
nn.Conv2d(in_channels=32, out_channels=32, kernel_size=3, padding=1),
|
35 |
+
nn.BatchNorm2d(32),
|
36 |
+
nn.ReLU(),
|
37 |
+
nn.Conv2d(in_channels=32, out_channels=64, kernel_size=4, stride=2, padding=1),
|
38 |
+
nn.BatchNorm2d(64),
|
39 |
+
nn.ReLU(),
|
40 |
+
|
41 |
+
nn.Conv2d(in_channels=64, out_channels=64, kernel_size=3, padding=1),
|
42 |
+
nn.BatchNorm2d(64),
|
43 |
+
nn.ReLU(),
|
44 |
+
nn.Conv2d(in_channels=64, out_channels=128, kernel_size=3, stride=1, padding=1),
|
45 |
+
nn.BatchNorm2d(128),
|
46 |
+
nn.ReLU()
|
47 |
+
)
|
48 |
+
|
49 |
+
# Final projection layer
|
50 |
+
self.final_proj = nn.Conv2d(in_channels=128, out_channels=noise_latent_channels, kernel_size=1)
|
51 |
+
|
52 |
+
self.conv_layers_1 = nn.Sequential(
|
53 |
+
nn.Conv2d(in_channels=noise_latent_channels, out_channels=noise_latent_channels, kernel_size=3, padding=1),
|
54 |
+
nn.BatchNorm2d(noise_latent_channels),
|
55 |
+
nn.ReLU(),
|
56 |
+
nn.Conv2d(in_channels=noise_latent_channels, out_channels=noise_latent_channels, kernel_size=3, stride=2, padding=1),
|
57 |
+
nn.BatchNorm2d(noise_latent_channels),
|
58 |
+
nn.ReLU(),
|
59 |
+
)
|
60 |
+
|
61 |
+
self.conv_layers_2 = nn.Sequential(
|
62 |
+
nn.Conv2d(in_channels=noise_latent_channels, out_channels=noise_latent_channels, kernel_size=3, padding=1),
|
63 |
+
nn.BatchNorm2d(noise_latent_channels),
|
64 |
+
nn.ReLU(),
|
65 |
+
nn.Conv2d(in_channels=noise_latent_channels, out_channels=noise_latent_channels*2, kernel_size=3, stride=2, padding=1),
|
66 |
+
nn.BatchNorm2d(noise_latent_channels*2),
|
67 |
+
nn.ReLU(),
|
68 |
+
)
|
69 |
+
|
70 |
+
self.conv_layers_3 = nn.Sequential(
|
71 |
+
nn.Conv2d(in_channels=noise_latent_channels*2, out_channels=noise_latent_channels*2, kernel_size=3, padding=1),
|
72 |
+
nn.BatchNorm2d(noise_latent_channels*2),
|
73 |
+
nn.ReLU(),
|
74 |
+
nn.Conv2d(in_channels=noise_latent_channels*2, out_channels=noise_latent_channels*4, kernel_size=3, stride=2, padding=1),
|
75 |
+
nn.BatchNorm2d(noise_latent_channels*4),
|
76 |
+
nn.ReLU(),
|
77 |
+
)
|
78 |
+
|
79 |
+
self.conv_layers_4 = nn.Sequential(
|
80 |
+
nn.Conv2d(in_channels=noise_latent_channels*4, out_channels=noise_latent_channels*4, kernel_size=3, padding=1),
|
81 |
+
nn.BatchNorm2d(noise_latent_channels*4),
|
82 |
+
nn.ReLU(),
|
83 |
+
)
|
84 |
+
|
85 |
+
if self.use_ca:
|
86 |
+
self.cross_attn1 = Transformer2DModel(in_channels=noise_latent_channels)
|
87 |
+
self.cross_attn2 = Transformer2DModel(in_channels=noise_latent_channels*2)
|
88 |
+
self.cross_attn3 = Transformer2DModel(in_channels=noise_latent_channels*4)
|
89 |
+
self.cross_attn4 = Transformer2DModel(in_channels=noise_latent_channels*4)
|
90 |
+
|
91 |
+
# Initialize layers
|
92 |
+
self._initialize_weights()
|
93 |
+
|
94 |
+
self.scale = nn.Parameter(torch.ones(1) * 2)
|
95 |
+
|
96 |
+
# def _initialize_weights(self):
|
97 |
+
# # Initialize weights with Gaussian distribution and zero out the final layer
|
98 |
+
# for m in self.conv_layers:
|
99 |
+
# if isinstance(m, nn.Conv2d):
|
100 |
+
# init.normal_(m.weight, mean=0.0, std=0.02)
|
101 |
+
# if m.bias is not None:
|
102 |
+
# init.zeros_(m.bias)
|
103 |
+
|
104 |
+
# init.zeros_(self.final_proj.weight)
|
105 |
+
# if self.final_proj.bias is not None:
|
106 |
+
# init.zeros_(self.final_proj.bias)
|
107 |
+
|
108 |
+
def _initialize_weights(self):
|
109 |
+
# Initialize weights with He initialization and zero out the biases
|
110 |
+
conv_blocks = [self.conv_layers, self.conv_layers_1, self.conv_layers_2, self.conv_layers_3, self.conv_layers_4]
|
111 |
+
for block_item in conv_blocks:
|
112 |
+
for m in block_item:
|
113 |
+
if isinstance(m, nn.Conv2d):
|
114 |
+
n = m.kernel_size[0] * m.kernel_size[1] * m.in_channels
|
115 |
+
init.normal_(m.weight, mean=0.0, std=np.sqrt(2. / n))
|
116 |
+
if m.bias is not None:
|
117 |
+
init.zeros_(m.bias)
|
118 |
+
|
119 |
+
# For the final projection layer, initialize weights to zero (or you may choose to use He initialization here as well)
|
120 |
+
init.zeros_(self.final_proj.weight)
|
121 |
+
if self.final_proj.bias is not None:
|
122 |
+
init.zeros_(self.final_proj.bias)
|
123 |
+
|
124 |
+
def forward(self, x, ref_x):
|
125 |
+
fea = []
|
126 |
+
b = x.shape[0]
|
127 |
+
|
128 |
+
x = rearrange(x, "b c f h w -> (b f) c h w")
|
129 |
+
x = self.conv_layers(x)
|
130 |
+
x = self.final_proj(x)
|
131 |
+
x = x * self.scale
|
132 |
+
# x = rearrange(x, "(b f) c h w -> b c f h w", b=b)
|
133 |
+
fea.append(rearrange(x, "(b f) c h w -> b c f h w", b=b))
|
134 |
+
|
135 |
+
x = self.conv_layers_1(x)
|
136 |
+
if self.use_ca:
|
137 |
+
ref_x = self.conv_layers(ref_x)
|
138 |
+
ref_x = self.final_proj(ref_x)
|
139 |
+
ref_x = ref_x * self.scale
|
140 |
+
ref_x = self.conv_layers_1(ref_x)
|
141 |
+
x = self.cross_attn1(x, ref_x)
|
142 |
+
fea.append(rearrange(x, "(b f) c h w -> b c f h w", b=b))
|
143 |
+
|
144 |
+
x = self.conv_layers_2(x)
|
145 |
+
if self.use_ca:
|
146 |
+
ref_x = self.conv_layers_2(ref_x)
|
147 |
+
x = self.cross_attn2(x, ref_x)
|
148 |
+
fea.append(rearrange(x, "(b f) c h w -> b c f h w", b=b))
|
149 |
+
|
150 |
+
x = self.conv_layers_3(x)
|
151 |
+
if self.use_ca:
|
152 |
+
ref_x = self.conv_layers_3(ref_x)
|
153 |
+
x = self.cross_attn3(x, ref_x)
|
154 |
+
fea.append(rearrange(x, "(b f) c h w -> b c f h w", b=b))
|
155 |
+
|
156 |
+
x = self.conv_layers_4(x)
|
157 |
+
if self.use_ca:
|
158 |
+
ref_x = self.conv_layers_4(ref_x)
|
159 |
+
x = self.cross_attn4(x, ref_x)
|
160 |
+
fea.append(rearrange(x, "(b f) c h w -> b c f h w", b=b))
|
161 |
+
|
162 |
+
return fea
|
163 |
+
|
164 |
+
# @classmethod
|
165 |
+
# def from_pretrained(cls,pretrained_model_path):
|
166 |
+
# if not os.path.exists(pretrained_model_path):
|
167 |
+
# print(f"There is no model file in {pretrained_model_path}")
|
168 |
+
# print(f"loaded PoseGuider's pretrained weights from {pretrained_model_path} ...")
|
169 |
+
|
170 |
+
# state_dict = torch.load(pretrained_model_path, map_location="cpu")
|
171 |
+
# model = Hack_PoseGuider(noise_latent_channels=320)
|
172 |
+
|
173 |
+
# m, u = model.load_state_dict(state_dict, strict=True)
|
174 |
+
# # print(f"### missing keys: {len(m)}; \n### unexpected keys: {len(u)};")
|
175 |
+
# params = [p.numel() for n, p in model.named_parameters()]
|
176 |
+
# print(f"### PoseGuider's Parameters: {sum(params) / 1e6} M")
|
177 |
+
|
178 |
+
# return model
|
179 |
+
|
180 |
+
|
181 |
+
class Transformer2DModel(ModelMixin):
|
182 |
+
_supports_gradient_checkpointing = True
|
183 |
+
def __init__(
|
184 |
+
self,
|
185 |
+
num_attention_heads: int = 16,
|
186 |
+
attention_head_dim: int = 88,
|
187 |
+
in_channels: Optional[int] = None,
|
188 |
+
num_layers: int = 1,
|
189 |
+
dropout: float = 0.0,
|
190 |
+
norm_num_groups: int = 32,
|
191 |
+
cross_attention_dim: Optional[int] = None,
|
192 |
+
attention_bias: bool = False,
|
193 |
+
activation_fn: str = "geglu",
|
194 |
+
num_embeds_ada_norm: Optional[int] = None,
|
195 |
+
use_linear_projection: bool = False,
|
196 |
+
only_cross_attention: bool = False,
|
197 |
+
double_self_attention: bool = False,
|
198 |
+
upcast_attention: bool = False,
|
199 |
+
norm_type: str = "layer_norm",
|
200 |
+
norm_elementwise_affine: bool = True,
|
201 |
+
norm_eps: float = 1e-5,
|
202 |
+
attention_type: str = "default",
|
203 |
+
):
|
204 |
+
super().__init__()
|
205 |
+
self.use_linear_projection = use_linear_projection
|
206 |
+
self.num_attention_heads = num_attention_heads
|
207 |
+
self.attention_head_dim = attention_head_dim
|
208 |
+
inner_dim = num_attention_heads * attention_head_dim
|
209 |
+
|
210 |
+
self.in_channels = in_channels
|
211 |
+
|
212 |
+
self.norm = torch.nn.GroupNorm(
|
213 |
+
num_groups=norm_num_groups,
|
214 |
+
num_channels=in_channels,
|
215 |
+
eps=1e-6,
|
216 |
+
affine=True,
|
217 |
+
)
|
218 |
+
if use_linear_projection:
|
219 |
+
self.proj_in = nn.Linear(in_channels, inner_dim)
|
220 |
+
else:
|
221 |
+
self.proj_in = nn.Conv2d(
|
222 |
+
in_channels, inner_dim, kernel_size=1, stride=1, padding=0
|
223 |
+
)
|
224 |
+
|
225 |
+
# 3. Define transformers blocks
|
226 |
+
self.transformer_blocks = nn.ModuleList(
|
227 |
+
[
|
228 |
+
BasicTransformerBlock(
|
229 |
+
inner_dim,
|
230 |
+
num_attention_heads,
|
231 |
+
attention_head_dim,
|
232 |
+
dropout=dropout,
|
233 |
+
cross_attention_dim=cross_attention_dim,
|
234 |
+
activation_fn=activation_fn,
|
235 |
+
num_embeds_ada_norm=num_embeds_ada_norm,
|
236 |
+
attention_bias=attention_bias,
|
237 |
+
only_cross_attention=only_cross_attention,
|
238 |
+
double_self_attention=double_self_attention,
|
239 |
+
upcast_attention=upcast_attention,
|
240 |
+
norm_type=norm_type,
|
241 |
+
norm_elementwise_affine=norm_elementwise_affine,
|
242 |
+
norm_eps=norm_eps,
|
243 |
+
attention_type=attention_type,
|
244 |
+
)
|
245 |
+
for d in range(num_layers)
|
246 |
+
]
|
247 |
+
)
|
248 |
+
|
249 |
+
if use_linear_projection:
|
250 |
+
self.proj_out = nn.Linear(inner_dim, in_channels)
|
251 |
+
else:
|
252 |
+
self.proj_out = nn.Conv2d(
|
253 |
+
inner_dim, in_channels, kernel_size=1, stride=1, padding=0
|
254 |
+
)
|
255 |
+
|
256 |
+
self.gradient_checkpointing = False
|
257 |
+
|
258 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
259 |
+
if hasattr(module, "gradient_checkpointing"):
|
260 |
+
module.gradient_checkpointing = value
|
261 |
+
|
262 |
+
def forward(
|
263 |
+
self,
|
264 |
+
hidden_states: torch.Tensor,
|
265 |
+
encoder_hidden_states: Optional[torch.Tensor] = None,
|
266 |
+
timestep: Optional[torch.LongTensor] = None,
|
267 |
+
):
|
268 |
+
batch, _, height, width = hidden_states.shape
|
269 |
+
residual = hidden_states
|
270 |
+
|
271 |
+
hidden_states = self.norm(hidden_states)
|
272 |
+
if not self.use_linear_projection:
|
273 |
+
hidden_states = self.proj_in(hidden_states)
|
274 |
+
inner_dim = hidden_states.shape[1]
|
275 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(
|
276 |
+
batch, height * width, inner_dim
|
277 |
+
)
|
278 |
+
else:
|
279 |
+
inner_dim = hidden_states.shape[1]
|
280 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(
|
281 |
+
batch, height * width, inner_dim
|
282 |
+
)
|
283 |
+
hidden_states = self.proj_in(hidden_states)
|
284 |
+
|
285 |
+
for block in self.transformer_blocks:
|
286 |
+
hidden_states = block(
|
287 |
+
hidden_states,
|
288 |
+
encoder_hidden_states=encoder_hidden_states,
|
289 |
+
timestep=timestep,
|
290 |
+
)
|
291 |
+
|
292 |
+
if not self.use_linear_projection:
|
293 |
+
hidden_states = (
|
294 |
+
hidden_states.reshape(batch, height, width, inner_dim)
|
295 |
+
.permute(0, 3, 1, 2)
|
296 |
+
.contiguous()
|
297 |
+
)
|
298 |
+
hidden_states = self.proj_out(hidden_states)
|
299 |
+
else:
|
300 |
+
hidden_states = self.proj_out(hidden_states)
|
301 |
+
hidden_states = (
|
302 |
+
hidden_states.reshape(batch, height, width, inner_dim)
|
303 |
+
.permute(0, 3, 1, 2)
|
304 |
+
.contiguous()
|
305 |
+
)
|
306 |
+
|
307 |
+
output = hidden_states + residual
|
308 |
+
return output
|
309 |
+
|
310 |
+
|
311 |
+
if __name__ == '__main__':
|
312 |
+
model = PoseGuider(noise_latent_channels=320).to(device="cuda")
|
313 |
+
|
314 |
+
input_data = torch.randn(1,3,1,512,512).to(device="cuda")
|
315 |
+
input_data1 = torch.randn(1,3,512,512).to(device="cuda")
|
316 |
+
|
317 |
+
output = model(input_data, input_data1)
|
318 |
+
for item in output:
|
319 |
+
print(item.shape)
|
320 |
+
|
321 |
+
# tf_model = Transformer2DModel(
|
322 |
+
# in_channels=320
|
323 |
+
# ).to('cuda')
|
324 |
+
|
325 |
+
# input_data = torch.randn(4,320,32,32).to(device="cuda")
|
326 |
+
# # input_emb = torch.randn(4,1,768).to(device="cuda")
|
327 |
+
# input_emb = torch.randn(4,320,32,32).to(device="cuda")
|
328 |
+
# o1 = tf_model(input_data, input_emb)
|
329 |
+
# print(o1.shape)
|
src/models/resnet.py
ADDED
@@ -0,0 +1,252 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/resnet.py
|
2 |
+
|
3 |
+
import torch
|
4 |
+
import torch.nn as nn
|
5 |
+
import torch.nn.functional as F
|
6 |
+
from einops import rearrange
|
7 |
+
from typing import Dict, Optional
|
8 |
+
|
9 |
+
|
10 |
+
class InflatedConv3d(nn.Conv2d):
|
11 |
+
def forward(self, x):
|
12 |
+
video_length = x.shape[2]
|
13 |
+
|
14 |
+
x = rearrange(x, "b c f h w -> (b f) c h w")
|
15 |
+
x = super().forward(x)
|
16 |
+
x = rearrange(x, "(b f) c h w -> b c f h w", f=video_length)
|
17 |
+
|
18 |
+
return x
|
19 |
+
|
20 |
+
|
21 |
+
class InflatedGroupNorm(nn.GroupNorm):
|
22 |
+
def forward(self, x):
|
23 |
+
video_length = x.shape[2]
|
24 |
+
|
25 |
+
x = rearrange(x, "b c f h w -> (b f) c h w")
|
26 |
+
x = super().forward(x)
|
27 |
+
x = rearrange(x, "(b f) c h w -> b c f h w", f=video_length)
|
28 |
+
|
29 |
+
return x
|
30 |
+
|
31 |
+
|
32 |
+
class Upsample3D(nn.Module):
|
33 |
+
def __init__(
|
34 |
+
self,
|
35 |
+
channels,
|
36 |
+
use_conv=False,
|
37 |
+
use_conv_transpose=False,
|
38 |
+
out_channels=None,
|
39 |
+
name="conv",
|
40 |
+
):
|
41 |
+
super().__init__()
|
42 |
+
self.channels = channels
|
43 |
+
self.out_channels = out_channels or channels
|
44 |
+
self.use_conv = use_conv
|
45 |
+
self.use_conv_transpose = use_conv_transpose
|
46 |
+
self.name = name
|
47 |
+
|
48 |
+
conv = None
|
49 |
+
if use_conv_transpose:
|
50 |
+
raise NotImplementedError
|
51 |
+
elif use_conv:
|
52 |
+
self.conv = InflatedConv3d(self.channels, self.out_channels, 3, padding=1)
|
53 |
+
|
54 |
+
def forward(self, hidden_states, output_size=None):
|
55 |
+
assert hidden_states.shape[1] == self.channels
|
56 |
+
|
57 |
+
if self.use_conv_transpose:
|
58 |
+
raise NotImplementedError
|
59 |
+
|
60 |
+
# Cast to float32 to as 'upsample_nearest2d_out_frame' op does not support bfloat16
|
61 |
+
dtype = hidden_states.dtype
|
62 |
+
if dtype == torch.bfloat16:
|
63 |
+
hidden_states = hidden_states.to(torch.float32)
|
64 |
+
|
65 |
+
# upsample_nearest_nhwc fails with large batch sizes. see https://github.com/huggingface/diffusers/issues/984
|
66 |
+
if hidden_states.shape[0] >= 64:
|
67 |
+
hidden_states = hidden_states.contiguous()
|
68 |
+
|
69 |
+
# if `output_size` is passed we force the interpolation output
|
70 |
+
# size and do not make use of `scale_factor=2`
|
71 |
+
if output_size is None:
|
72 |
+
hidden_states = F.interpolate(
|
73 |
+
hidden_states, scale_factor=[1.0, 2.0, 2.0], mode="nearest"
|
74 |
+
)
|
75 |
+
else:
|
76 |
+
hidden_states = F.interpolate(
|
77 |
+
hidden_states, size=output_size, mode="nearest"
|
78 |
+
)
|
79 |
+
|
80 |
+
# If the input is bfloat16, we cast back to bfloat16
|
81 |
+
if dtype == torch.bfloat16:
|
82 |
+
hidden_states = hidden_states.to(dtype)
|
83 |
+
|
84 |
+
# if self.use_conv:
|
85 |
+
# if self.name == "conv":
|
86 |
+
# hidden_states = self.conv(hidden_states)
|
87 |
+
# else:
|
88 |
+
# hidden_states = self.Conv2d_0(hidden_states)
|
89 |
+
hidden_states = self.conv(hidden_states)
|
90 |
+
|
91 |
+
return hidden_states
|
92 |
+
|
93 |
+
|
94 |
+
class Downsample3D(nn.Module):
|
95 |
+
def __init__(
|
96 |
+
self, channels, use_conv=False, out_channels=None, padding=1, name="conv"
|
97 |
+
):
|
98 |
+
super().__init__()
|
99 |
+
self.channels = channels
|
100 |
+
self.out_channels = out_channels or channels
|
101 |
+
self.use_conv = use_conv
|
102 |
+
self.padding = padding
|
103 |
+
stride = 2
|
104 |
+
self.name = name
|
105 |
+
|
106 |
+
if use_conv:
|
107 |
+
self.conv = InflatedConv3d(
|
108 |
+
self.channels, self.out_channels, 3, stride=stride, padding=padding
|
109 |
+
)
|
110 |
+
else:
|
111 |
+
raise NotImplementedError
|
112 |
+
|
113 |
+
def forward(self, hidden_states):
|
114 |
+
assert hidden_states.shape[1] == self.channels
|
115 |
+
if self.use_conv and self.padding == 0:
|
116 |
+
raise NotImplementedError
|
117 |
+
|
118 |
+
assert hidden_states.shape[1] == self.channels
|
119 |
+
hidden_states = self.conv(hidden_states)
|
120 |
+
|
121 |
+
return hidden_states
|
122 |
+
|
123 |
+
|
124 |
+
class ResnetBlock3D(nn.Module):
|
125 |
+
def __init__(
|
126 |
+
self,
|
127 |
+
*,
|
128 |
+
in_channels,
|
129 |
+
out_channels=None,
|
130 |
+
conv_shortcut=False,
|
131 |
+
dropout=0.0,
|
132 |
+
temb_channels=512,
|
133 |
+
groups=32,
|
134 |
+
groups_out=None,
|
135 |
+
pre_norm=True,
|
136 |
+
eps=1e-6,
|
137 |
+
non_linearity="swish",
|
138 |
+
time_embedding_norm="default",
|
139 |
+
output_scale_factor=1.0,
|
140 |
+
use_in_shortcut=None,
|
141 |
+
use_inflated_groupnorm=None,
|
142 |
+
):
|
143 |
+
super().__init__()
|
144 |
+
self.pre_norm = pre_norm
|
145 |
+
self.pre_norm = True
|
146 |
+
self.in_channels = in_channels
|
147 |
+
out_channels = in_channels if out_channels is None else out_channels
|
148 |
+
self.out_channels = out_channels
|
149 |
+
self.use_conv_shortcut = conv_shortcut
|
150 |
+
self.time_embedding_norm = time_embedding_norm
|
151 |
+
self.output_scale_factor = output_scale_factor
|
152 |
+
|
153 |
+
if groups_out is None:
|
154 |
+
groups_out = groups
|
155 |
+
|
156 |
+
assert use_inflated_groupnorm != None
|
157 |
+
if use_inflated_groupnorm:
|
158 |
+
self.norm1 = InflatedGroupNorm(
|
159 |
+
num_groups=groups, num_channels=in_channels, eps=eps, affine=True
|
160 |
+
)
|
161 |
+
else:
|
162 |
+
self.norm1 = torch.nn.GroupNorm(
|
163 |
+
num_groups=groups, num_channels=in_channels, eps=eps, affine=True
|
164 |
+
)
|
165 |
+
|
166 |
+
self.conv1 = InflatedConv3d(
|
167 |
+
in_channels, out_channels, kernel_size=3, stride=1, padding=1
|
168 |
+
)
|
169 |
+
|
170 |
+
if temb_channels is not None:
|
171 |
+
if self.time_embedding_norm == "default":
|
172 |
+
time_emb_proj_out_channels = out_channels
|
173 |
+
elif self.time_embedding_norm == "scale_shift":
|
174 |
+
time_emb_proj_out_channels = out_channels * 2
|
175 |
+
else:
|
176 |
+
raise ValueError(
|
177 |
+
f"unknown time_embedding_norm : {self.time_embedding_norm} "
|
178 |
+
)
|
179 |
+
|
180 |
+
self.time_emb_proj = torch.nn.Linear(
|
181 |
+
temb_channels, time_emb_proj_out_channels
|
182 |
+
)
|
183 |
+
else:
|
184 |
+
self.time_emb_proj = None
|
185 |
+
|
186 |
+
if use_inflated_groupnorm:
|
187 |
+
self.norm2 = InflatedGroupNorm(
|
188 |
+
num_groups=groups_out, num_channels=out_channels, eps=eps, affine=True
|
189 |
+
)
|
190 |
+
else:
|
191 |
+
self.norm2 = torch.nn.GroupNorm(
|
192 |
+
num_groups=groups_out, num_channels=out_channels, eps=eps, affine=True
|
193 |
+
)
|
194 |
+
self.dropout = torch.nn.Dropout(dropout)
|
195 |
+
self.conv2 = InflatedConv3d(
|
196 |
+
out_channels, out_channels, kernel_size=3, stride=1, padding=1
|
197 |
+
)
|
198 |
+
|
199 |
+
if non_linearity == "swish":
|
200 |
+
self.nonlinearity = lambda x: F.silu(x)
|
201 |
+
elif non_linearity == "mish":
|
202 |
+
self.nonlinearity = Mish()
|
203 |
+
elif non_linearity == "silu":
|
204 |
+
self.nonlinearity = nn.SiLU()
|
205 |
+
|
206 |
+
self.use_in_shortcut = (
|
207 |
+
self.in_channels != self.out_channels
|
208 |
+
if use_in_shortcut is None
|
209 |
+
else use_in_shortcut
|
210 |
+
)
|
211 |
+
|
212 |
+
self.conv_shortcut = None
|
213 |
+
if self.use_in_shortcut:
|
214 |
+
self.conv_shortcut = InflatedConv3d(
|
215 |
+
in_channels, out_channels, kernel_size=1, stride=1, padding=0
|
216 |
+
)
|
217 |
+
|
218 |
+
def forward(self, input_tensor, temb):
|
219 |
+
hidden_states = input_tensor
|
220 |
+
|
221 |
+
hidden_states = self.norm1(hidden_states)
|
222 |
+
hidden_states = self.nonlinearity(hidden_states)
|
223 |
+
|
224 |
+
hidden_states = self.conv1(hidden_states)
|
225 |
+
|
226 |
+
if temb is not None:
|
227 |
+
temb = self.time_emb_proj(self.nonlinearity(temb))[:, :, None, None, None]
|
228 |
+
|
229 |
+
if temb is not None and self.time_embedding_norm == "default":
|
230 |
+
hidden_states = hidden_states + temb
|
231 |
+
|
232 |
+
hidden_states = self.norm2(hidden_states)
|
233 |
+
|
234 |
+
if temb is not None and self.time_embedding_norm == "scale_shift":
|
235 |
+
scale, shift = torch.chunk(temb, 2, dim=1)
|
236 |
+
hidden_states = hidden_states * (1 + scale) + shift
|
237 |
+
|
238 |
+
hidden_states = self.nonlinearity(hidden_states)
|
239 |
+
|
240 |
+
hidden_states = self.dropout(hidden_states)
|
241 |
+
hidden_states = self.conv2(hidden_states)
|
242 |
+
|
243 |
+
if self.conv_shortcut is not None:
|
244 |
+
input_tensor = self.conv_shortcut(input_tensor)
|
245 |
+
|
246 |
+
output_tensor = (input_tensor + hidden_states) / self.output_scale_factor
|
247 |
+
|
248 |
+
return output_tensor
|
249 |
+
|
250 |
+
class Mish(torch.nn.Module):
|
251 |
+
def forward(self, hidden_states):
|
252 |
+
return hidden_states * torch.tanh(torch.nn.functional.softplus(hidden_states))
|
src/models/transformer_2d.py
ADDED
@@ -0,0 +1,396 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/transformer_2d.py
|
2 |
+
from dataclasses import dataclass
|
3 |
+
from typing import Any, Dict, Optional
|
4 |
+
|
5 |
+
import torch
|
6 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
7 |
+
from diffusers.models.embeddings import CaptionProjection
|
8 |
+
from diffusers.models.lora import LoRACompatibleConv, LoRACompatibleLinear
|
9 |
+
from diffusers.models.modeling_utils import ModelMixin
|
10 |
+
from diffusers.models.normalization import AdaLayerNormSingle
|
11 |
+
from diffusers.utils import USE_PEFT_BACKEND, BaseOutput, deprecate, is_torch_version
|
12 |
+
from torch import nn
|
13 |
+
|
14 |
+
from .attention import BasicTransformerBlock
|
15 |
+
|
16 |
+
|
17 |
+
@dataclass
|
18 |
+
class Transformer2DModelOutput(BaseOutput):
|
19 |
+
"""
|
20 |
+
The output of [`Transformer2DModel`].
|
21 |
+
|
22 |
+
Args:
|
23 |
+
sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)` or `(batch size, num_vector_embeds - 1, num_latent_pixels)` if [`Transformer2DModel`] is discrete):
|
24 |
+
The hidden states output conditioned on the `encoder_hidden_states` input. If discrete, returns probability
|
25 |
+
distributions for the unnoised latent pixels.
|
26 |
+
"""
|
27 |
+
|
28 |
+
sample: torch.FloatTensor
|
29 |
+
ref_feature: torch.FloatTensor
|
30 |
+
|
31 |
+
|
32 |
+
class Transformer2DModel(ModelMixin, ConfigMixin):
|
33 |
+
"""
|
34 |
+
A 2D Transformer model for image-like data.
|
35 |
+
|
36 |
+
Parameters:
|
37 |
+
num_attention_heads (`int`, *optional*, defaults to 16): The number of heads to use for multi-head attention.
|
38 |
+
attention_head_dim (`int`, *optional*, defaults to 88): The number of channels in each head.
|
39 |
+
in_channels (`int`, *optional*):
|
40 |
+
The number of channels in the input and output (specify if the input is **continuous**).
|
41 |
+
num_layers (`int`, *optional*, defaults to 1): The number of layers of Transformer blocks to use.
|
42 |
+
dropout (`float`, *optional*, defaults to 0.0): The dropout probability to use.
|
43 |
+
cross_attention_dim (`int`, *optional*): The number of `encoder_hidden_states` dimensions to use.
|
44 |
+
sample_size (`int`, *optional*): The width of the latent images (specify if the input is **discrete**).
|
45 |
+
This is fixed during training since it is used to learn a number of position embeddings.
|
46 |
+
num_vector_embeds (`int`, *optional*):
|
47 |
+
The number of classes of the vector embeddings of the latent pixels (specify if the input is **discrete**).
|
48 |
+
Includes the class for the masked latent pixel.
|
49 |
+
activation_fn (`str`, *optional*, defaults to `"geglu"`): Activation function to use in feed-forward.
|
50 |
+
num_embeds_ada_norm ( `int`, *optional*):
|
51 |
+
The number of diffusion steps used during training. Pass if at least one of the norm_layers is
|
52 |
+
`AdaLayerNorm`. This is fixed during training since it is used to learn a number of embeddings that are
|
53 |
+
added to the hidden states.
|
54 |
+
|
55 |
+
During inference, you can denoise for up to but not more steps than `num_embeds_ada_norm`.
|
56 |
+
attention_bias (`bool`, *optional*):
|
57 |
+
Configure if the `TransformerBlocks` attention should contain a bias parameter.
|
58 |
+
"""
|
59 |
+
|
60 |
+
_supports_gradient_checkpointing = True
|
61 |
+
|
62 |
+
@register_to_config
|
63 |
+
def __init__(
|
64 |
+
self,
|
65 |
+
num_attention_heads: int = 16,
|
66 |
+
attention_head_dim: int = 88,
|
67 |
+
in_channels: Optional[int] = None,
|
68 |
+
out_channels: Optional[int] = None,
|
69 |
+
num_layers: int = 1,
|
70 |
+
dropout: float = 0.0,
|
71 |
+
norm_num_groups: int = 32,
|
72 |
+
cross_attention_dim: Optional[int] = None,
|
73 |
+
attention_bias: bool = False,
|
74 |
+
sample_size: Optional[int] = None,
|
75 |
+
num_vector_embeds: Optional[int] = None,
|
76 |
+
patch_size: Optional[int] = None,
|
77 |
+
activation_fn: str = "geglu",
|
78 |
+
num_embeds_ada_norm: Optional[int] = None,
|
79 |
+
use_linear_projection: bool = False,
|
80 |
+
only_cross_attention: bool = False,
|
81 |
+
double_self_attention: bool = False,
|
82 |
+
upcast_attention: bool = False,
|
83 |
+
norm_type: str = "layer_norm",
|
84 |
+
norm_elementwise_affine: bool = True,
|
85 |
+
norm_eps: float = 1e-5,
|
86 |
+
attention_type: str = "default",
|
87 |
+
caption_channels: int = None,
|
88 |
+
):
|
89 |
+
super().__init__()
|
90 |
+
self.use_linear_projection = use_linear_projection
|
91 |
+
self.num_attention_heads = num_attention_heads
|
92 |
+
self.attention_head_dim = attention_head_dim
|
93 |
+
inner_dim = num_attention_heads * attention_head_dim
|
94 |
+
|
95 |
+
conv_cls = nn.Conv2d if USE_PEFT_BACKEND else LoRACompatibleConv
|
96 |
+
linear_cls = nn.Linear if USE_PEFT_BACKEND else LoRACompatibleLinear
|
97 |
+
|
98 |
+
# 1. Transformer2DModel can process both standard continuous images of shape `(batch_size, num_channels, width, height)` as well as quantized image embeddings of shape `(batch_size, num_image_vectors)`
|
99 |
+
# Define whether input is continuous or discrete depending on configuration
|
100 |
+
self.is_input_continuous = (in_channels is not None) and (patch_size is None)
|
101 |
+
self.is_input_vectorized = num_vector_embeds is not None
|
102 |
+
self.is_input_patches = in_channels is not None and patch_size is not None
|
103 |
+
|
104 |
+
if norm_type == "layer_norm" and num_embeds_ada_norm is not None:
|
105 |
+
deprecation_message = (
|
106 |
+
f"The configuration file of this model: {self.__class__} is outdated. `norm_type` is either not set or"
|
107 |
+
" incorrectly set to `'layer_norm'`.Make sure to set `norm_type` to `'ada_norm'` in the config."
|
108 |
+
" Please make sure to update the config accordingly as leaving `norm_type` might led to incorrect"
|
109 |
+
" results in future versions. If you have downloaded this checkpoint from the Hugging Face Hub, it"
|
110 |
+
" would be very nice if you could open a Pull request for the `transformer/config.json` file"
|
111 |
+
)
|
112 |
+
deprecate(
|
113 |
+
"norm_type!=num_embeds_ada_norm",
|
114 |
+
"1.0.0",
|
115 |
+
deprecation_message,
|
116 |
+
standard_warn=False,
|
117 |
+
)
|
118 |
+
norm_type = "ada_norm"
|
119 |
+
|
120 |
+
if self.is_input_continuous and self.is_input_vectorized:
|
121 |
+
raise ValueError(
|
122 |
+
f"Cannot define both `in_channels`: {in_channels} and `num_vector_embeds`: {num_vector_embeds}. Make"
|
123 |
+
" sure that either `in_channels` or `num_vector_embeds` is None."
|
124 |
+
)
|
125 |
+
elif self.is_input_vectorized and self.is_input_patches:
|
126 |
+
raise ValueError(
|
127 |
+
f"Cannot define both `num_vector_embeds`: {num_vector_embeds} and `patch_size`: {patch_size}. Make"
|
128 |
+
" sure that either `num_vector_embeds` or `num_patches` is None."
|
129 |
+
)
|
130 |
+
elif (
|
131 |
+
not self.is_input_continuous
|
132 |
+
and not self.is_input_vectorized
|
133 |
+
and not self.is_input_patches
|
134 |
+
):
|
135 |
+
raise ValueError(
|
136 |
+
f"Has to define `in_channels`: {in_channels}, `num_vector_embeds`: {num_vector_embeds}, or patch_size:"
|
137 |
+
f" {patch_size}. Make sure that `in_channels`, `num_vector_embeds` or `num_patches` is not None."
|
138 |
+
)
|
139 |
+
|
140 |
+
# 2. Define input layers
|
141 |
+
self.in_channels = in_channels
|
142 |
+
|
143 |
+
self.norm = torch.nn.GroupNorm(
|
144 |
+
num_groups=norm_num_groups,
|
145 |
+
num_channels=in_channels,
|
146 |
+
eps=1e-6,
|
147 |
+
affine=True,
|
148 |
+
)
|
149 |
+
if use_linear_projection:
|
150 |
+
self.proj_in = linear_cls(in_channels, inner_dim)
|
151 |
+
else:
|
152 |
+
self.proj_in = conv_cls(
|
153 |
+
in_channels, inner_dim, kernel_size=1, stride=1, padding=0
|
154 |
+
)
|
155 |
+
|
156 |
+
# 3. Define transformers blocks
|
157 |
+
self.transformer_blocks = nn.ModuleList(
|
158 |
+
[
|
159 |
+
BasicTransformerBlock(
|
160 |
+
inner_dim,
|
161 |
+
num_attention_heads,
|
162 |
+
attention_head_dim,
|
163 |
+
dropout=dropout,
|
164 |
+
cross_attention_dim=cross_attention_dim,
|
165 |
+
activation_fn=activation_fn,
|
166 |
+
num_embeds_ada_norm=num_embeds_ada_norm,
|
167 |
+
attention_bias=attention_bias,
|
168 |
+
only_cross_attention=only_cross_attention,
|
169 |
+
double_self_attention=double_self_attention,
|
170 |
+
upcast_attention=upcast_attention,
|
171 |
+
norm_type=norm_type,
|
172 |
+
norm_elementwise_affine=norm_elementwise_affine,
|
173 |
+
norm_eps=norm_eps,
|
174 |
+
attention_type=attention_type,
|
175 |
+
)
|
176 |
+
for d in range(num_layers)
|
177 |
+
]
|
178 |
+
)
|
179 |
+
|
180 |
+
# 4. Define output layers
|
181 |
+
self.out_channels = in_channels if out_channels is None else out_channels
|
182 |
+
# TODO: should use out_channels for continuous projections
|
183 |
+
if use_linear_projection:
|
184 |
+
self.proj_out = linear_cls(inner_dim, in_channels)
|
185 |
+
else:
|
186 |
+
self.proj_out = conv_cls(
|
187 |
+
inner_dim, in_channels, kernel_size=1, stride=1, padding=0
|
188 |
+
)
|
189 |
+
|
190 |
+
# 5. PixArt-Alpha blocks.
|
191 |
+
self.adaln_single = None
|
192 |
+
self.use_additional_conditions = False
|
193 |
+
if norm_type == "ada_norm_single":
|
194 |
+
self.use_additional_conditions = self.config.sample_size == 128
|
195 |
+
# TODO(Sayak, PVP) clean this, for now we use sample size to determine whether to use
|
196 |
+
# additional conditions until we find better name
|
197 |
+
self.adaln_single = AdaLayerNormSingle(
|
198 |
+
inner_dim, use_additional_conditions=self.use_additional_conditions
|
199 |
+
)
|
200 |
+
|
201 |
+
self.caption_projection = None
|
202 |
+
if caption_channels is not None:
|
203 |
+
self.caption_projection = CaptionProjection(
|
204 |
+
in_features=caption_channels, hidden_size=inner_dim
|
205 |
+
)
|
206 |
+
|
207 |
+
self.gradient_checkpointing = False
|
208 |
+
|
209 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
210 |
+
if hasattr(module, "gradient_checkpointing"):
|
211 |
+
module.gradient_checkpointing = value
|
212 |
+
|
213 |
+
def forward(
|
214 |
+
self,
|
215 |
+
hidden_states: torch.Tensor,
|
216 |
+
encoder_hidden_states: Optional[torch.Tensor] = None,
|
217 |
+
timestep: Optional[torch.LongTensor] = None,
|
218 |
+
added_cond_kwargs: Dict[str, torch.Tensor] = None,
|
219 |
+
class_labels: Optional[torch.LongTensor] = None,
|
220 |
+
cross_attention_kwargs: Dict[str, Any] = None,
|
221 |
+
attention_mask: Optional[torch.Tensor] = None,
|
222 |
+
encoder_attention_mask: Optional[torch.Tensor] = None,
|
223 |
+
return_dict: bool = True,
|
224 |
+
):
|
225 |
+
"""
|
226 |
+
The [`Transformer2DModel`] forward method.
|
227 |
+
|
228 |
+
Args:
|
229 |
+
hidden_states (`torch.LongTensor` of shape `(batch size, num latent pixels)` if discrete, `torch.FloatTensor` of shape `(batch size, channel, height, width)` if continuous):
|
230 |
+
Input `hidden_states`.
|
231 |
+
encoder_hidden_states ( `torch.FloatTensor` of shape `(batch size, sequence len, embed dims)`, *optional*):
|
232 |
+
Conditional embeddings for cross attention layer. If not given, cross-attention defaults to
|
233 |
+
self-attention.
|
234 |
+
timestep ( `torch.LongTensor`, *optional*):
|
235 |
+
Used to indicate denoising step. Optional timestep to be applied as an embedding in `AdaLayerNorm`.
|
236 |
+
class_labels ( `torch.LongTensor` of shape `(batch size, num classes)`, *optional*):
|
237 |
+
Used to indicate class labels conditioning. Optional class labels to be applied as an embedding in
|
238 |
+
`AdaLayerZeroNorm`.
|
239 |
+
cross_attention_kwargs ( `Dict[str, Any]`, *optional*):
|
240 |
+
A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
|
241 |
+
`self.processor` in
|
242 |
+
[diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
|
243 |
+
attention_mask ( `torch.Tensor`, *optional*):
|
244 |
+
An attention mask of shape `(batch, key_tokens)` is applied to `encoder_hidden_states`. If `1` the mask
|
245 |
+
is kept, otherwise if `0` it is discarded. Mask will be converted into a bias, which adds large
|
246 |
+
negative values to the attention scores corresponding to "discard" tokens.
|
247 |
+
encoder_attention_mask ( `torch.Tensor`, *optional*):
|
248 |
+
Cross-attention mask applied to `encoder_hidden_states`. Two formats supported:
|
249 |
+
|
250 |
+
* Mask `(batch, sequence_length)` True = keep, False = discard.
|
251 |
+
* Bias `(batch, 1, sequence_length)` 0 = keep, -10000 = discard.
|
252 |
+
|
253 |
+
If `ndim == 2`: will be interpreted as a mask, then converted into a bias consistent with the format
|
254 |
+
above. This bias will be added to the cross-attention scores.
|
255 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
256 |
+
Whether or not to return a [`~models.unet_2d_condition.UNet2DConditionOutput`] instead of a plain
|
257 |
+
tuple.
|
258 |
+
|
259 |
+
Returns:
|
260 |
+
If `return_dict` is True, an [`~models.transformer_2d.Transformer2DModelOutput`] is returned, otherwise a
|
261 |
+
`tuple` where the first element is the sample tensor.
|
262 |
+
"""
|
263 |
+
# ensure attention_mask is a bias, and give it a singleton query_tokens dimension.
|
264 |
+
# we may have done this conversion already, e.g. if we came here via UNet2DConditionModel#forward.
|
265 |
+
# we can tell by counting dims; if ndim == 2: it's a mask rather than a bias.
|
266 |
+
# expects mask of shape:
|
267 |
+
# [batch, key_tokens]
|
268 |
+
# adds singleton query_tokens dimension:
|
269 |
+
# [batch, 1, key_tokens]
|
270 |
+
# this helps to broadcast it as a bias over attention scores, which will be in one of the following shapes:
|
271 |
+
# [batch, heads, query_tokens, key_tokens] (e.g. torch sdp attn)
|
272 |
+
# [batch * heads, query_tokens, key_tokens] (e.g. xformers or classic attn)
|
273 |
+
if attention_mask is not None and attention_mask.ndim == 2:
|
274 |
+
# assume that mask is expressed as:
|
275 |
+
# (1 = keep, 0 = discard)
|
276 |
+
# convert mask into a bias that can be added to attention scores:
|
277 |
+
# (keep = +0, discard = -10000.0)
|
278 |
+
attention_mask = (1 - attention_mask.to(hidden_states.dtype)) * -10000.0
|
279 |
+
attention_mask = attention_mask.unsqueeze(1)
|
280 |
+
|
281 |
+
# convert encoder_attention_mask to a bias the same way we do for attention_mask
|
282 |
+
if encoder_attention_mask is not None and encoder_attention_mask.ndim == 2:
|
283 |
+
encoder_attention_mask = (
|
284 |
+
1 - encoder_attention_mask.to(hidden_states.dtype)
|
285 |
+
) * -10000.0
|
286 |
+
encoder_attention_mask = encoder_attention_mask.unsqueeze(1)
|
287 |
+
|
288 |
+
# Retrieve lora scale.
|
289 |
+
lora_scale = (
|
290 |
+
cross_attention_kwargs.get("scale", 1.0)
|
291 |
+
if cross_attention_kwargs is not None
|
292 |
+
else 1.0
|
293 |
+
)
|
294 |
+
|
295 |
+
# 1. Input
|
296 |
+
batch, _, height, width = hidden_states.shape
|
297 |
+
residual = hidden_states
|
298 |
+
|
299 |
+
hidden_states = self.norm(hidden_states)
|
300 |
+
if not self.use_linear_projection:
|
301 |
+
hidden_states = (
|
302 |
+
self.proj_in(hidden_states, scale=lora_scale)
|
303 |
+
if not USE_PEFT_BACKEND
|
304 |
+
else self.proj_in(hidden_states)
|
305 |
+
)
|
306 |
+
inner_dim = hidden_states.shape[1]
|
307 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(
|
308 |
+
batch, height * width, inner_dim
|
309 |
+
)
|
310 |
+
else:
|
311 |
+
inner_dim = hidden_states.shape[1]
|
312 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(
|
313 |
+
batch, height * width, inner_dim
|
314 |
+
)
|
315 |
+
hidden_states = (
|
316 |
+
self.proj_in(hidden_states, scale=lora_scale)
|
317 |
+
if not USE_PEFT_BACKEND
|
318 |
+
else self.proj_in(hidden_states)
|
319 |
+
)
|
320 |
+
|
321 |
+
# 2. Blocks
|
322 |
+
if self.caption_projection is not None:
|
323 |
+
batch_size = hidden_states.shape[0]
|
324 |
+
encoder_hidden_states = self.caption_projection(encoder_hidden_states)
|
325 |
+
encoder_hidden_states = encoder_hidden_states.view(
|
326 |
+
batch_size, -1, hidden_states.shape[-1]
|
327 |
+
)
|
328 |
+
|
329 |
+
ref_feature = hidden_states.reshape(batch, height, width, inner_dim)
|
330 |
+
for block in self.transformer_blocks:
|
331 |
+
if self.training and self.gradient_checkpointing:
|
332 |
+
|
333 |
+
def create_custom_forward(module, return_dict=None):
|
334 |
+
def custom_forward(*inputs):
|
335 |
+
if return_dict is not None:
|
336 |
+
return module(*inputs, return_dict=return_dict)
|
337 |
+
else:
|
338 |
+
return module(*inputs)
|
339 |
+
|
340 |
+
return custom_forward
|
341 |
+
|
342 |
+
ckpt_kwargs: Dict[str, Any] = (
|
343 |
+
{"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
|
344 |
+
)
|
345 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
346 |
+
create_custom_forward(block),
|
347 |
+
hidden_states,
|
348 |
+
attention_mask,
|
349 |
+
encoder_hidden_states,
|
350 |
+
encoder_attention_mask,
|
351 |
+
timestep,
|
352 |
+
cross_attention_kwargs,
|
353 |
+
class_labels,
|
354 |
+
**ckpt_kwargs,
|
355 |
+
)
|
356 |
+
else:
|
357 |
+
hidden_states = block(
|
358 |
+
hidden_states,
|
359 |
+
attention_mask=attention_mask,
|
360 |
+
encoder_hidden_states=encoder_hidden_states,
|
361 |
+
encoder_attention_mask=encoder_attention_mask,
|
362 |
+
timestep=timestep,
|
363 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
364 |
+
class_labels=class_labels,
|
365 |
+
)
|
366 |
+
|
367 |
+
# 3. Output
|
368 |
+
if self.is_input_continuous:
|
369 |
+
if not self.use_linear_projection:
|
370 |
+
hidden_states = (
|
371 |
+
hidden_states.reshape(batch, height, width, inner_dim)
|
372 |
+
.permute(0, 3, 1, 2)
|
373 |
+
.contiguous()
|
374 |
+
)
|
375 |
+
hidden_states = (
|
376 |
+
self.proj_out(hidden_states, scale=lora_scale)
|
377 |
+
if not USE_PEFT_BACKEND
|
378 |
+
else self.proj_out(hidden_states)
|
379 |
+
)
|
380 |
+
else:
|
381 |
+
hidden_states = (
|
382 |
+
self.proj_out(hidden_states, scale=lora_scale)
|
383 |
+
if not USE_PEFT_BACKEND
|
384 |
+
else self.proj_out(hidden_states)
|
385 |
+
)
|
386 |
+
hidden_states = (
|
387 |
+
hidden_states.reshape(batch, height, width, inner_dim)
|
388 |
+
.permute(0, 3, 1, 2)
|
389 |
+
.contiguous()
|
390 |
+
)
|
391 |
+
|
392 |
+
output = hidden_states + residual
|
393 |
+
if not return_dict:
|
394 |
+
return (output, ref_feature)
|
395 |
+
|
396 |
+
return Transformer2DModelOutput(sample=output, ref_feature=ref_feature)
|
src/models/transformer_3d.py
ADDED
@@ -0,0 +1,169 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from dataclasses import dataclass
|
2 |
+
from typing import Optional, Dict
|
3 |
+
|
4 |
+
import torch
|
5 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
6 |
+
from diffusers.models import ModelMixin
|
7 |
+
from diffusers.utils import BaseOutput
|
8 |
+
from diffusers.utils.import_utils import is_xformers_available
|
9 |
+
from einops import rearrange, repeat
|
10 |
+
from torch import nn
|
11 |
+
|
12 |
+
from .attention import TemporalBasicTransformerBlock, ResidualTemporalBasicTransformerBlock
|
13 |
+
|
14 |
+
|
15 |
+
@dataclass
|
16 |
+
class Transformer3DModelOutput(BaseOutput):
|
17 |
+
sample: torch.FloatTensor
|
18 |
+
|
19 |
+
|
20 |
+
if is_xformers_available():
|
21 |
+
import xformers
|
22 |
+
import xformers.ops
|
23 |
+
else:
|
24 |
+
xformers = None
|
25 |
+
|
26 |
+
|
27 |
+
class Transformer3DModel(ModelMixin, ConfigMixin):
|
28 |
+
_supports_gradient_checkpointing = True
|
29 |
+
|
30 |
+
@register_to_config
|
31 |
+
def __init__(
|
32 |
+
self,
|
33 |
+
num_attention_heads: int = 16,
|
34 |
+
attention_head_dim: int = 88,
|
35 |
+
in_channels: Optional[int] = None,
|
36 |
+
num_layers: int = 1,
|
37 |
+
dropout: float = 0.0,
|
38 |
+
norm_num_groups: int = 32,
|
39 |
+
cross_attention_dim: Optional[int] = None,
|
40 |
+
attention_bias: bool = False,
|
41 |
+
activation_fn: str = "geglu",
|
42 |
+
num_embeds_ada_norm: Optional[int] = None,
|
43 |
+
use_linear_projection: bool = False,
|
44 |
+
only_cross_attention: bool = False,
|
45 |
+
upcast_attention: bool = False,
|
46 |
+
unet_use_cross_frame_attention=None,
|
47 |
+
unet_use_temporal_attention=None,
|
48 |
+
):
|
49 |
+
super().__init__()
|
50 |
+
self.use_linear_projection = use_linear_projection
|
51 |
+
self.num_attention_heads = num_attention_heads
|
52 |
+
self.attention_head_dim = attention_head_dim
|
53 |
+
inner_dim = num_attention_heads * attention_head_dim
|
54 |
+
|
55 |
+
# Define input layers
|
56 |
+
self.in_channels = in_channels
|
57 |
+
|
58 |
+
self.norm = torch.nn.GroupNorm(
|
59 |
+
num_groups=norm_num_groups, num_channels=in_channels, eps=1e-6, affine=True
|
60 |
+
)
|
61 |
+
if use_linear_projection:
|
62 |
+
self.proj_in = nn.Linear(in_channels, inner_dim)
|
63 |
+
else:
|
64 |
+
self.proj_in = nn.Conv2d(
|
65 |
+
in_channels, inner_dim, kernel_size=1, stride=1, padding=0
|
66 |
+
)
|
67 |
+
|
68 |
+
# Define transformers blocks
|
69 |
+
self.transformer_blocks = nn.ModuleList(
|
70 |
+
[
|
71 |
+
TemporalBasicTransformerBlock(
|
72 |
+
inner_dim,
|
73 |
+
num_attention_heads,
|
74 |
+
attention_head_dim,
|
75 |
+
dropout=dropout,
|
76 |
+
cross_attention_dim=cross_attention_dim,
|
77 |
+
activation_fn=activation_fn,
|
78 |
+
num_embeds_ada_norm=num_embeds_ada_norm,
|
79 |
+
attention_bias=attention_bias,
|
80 |
+
only_cross_attention=only_cross_attention,
|
81 |
+
upcast_attention=upcast_attention,
|
82 |
+
unet_use_cross_frame_attention=unet_use_cross_frame_attention,
|
83 |
+
unet_use_temporal_attention=unet_use_temporal_attention,
|
84 |
+
)
|
85 |
+
for d in range(num_layers)
|
86 |
+
]
|
87 |
+
)
|
88 |
+
|
89 |
+
# 4. Define output layers
|
90 |
+
if use_linear_projection:
|
91 |
+
self.proj_out = nn.Linear(in_channels, inner_dim)
|
92 |
+
else:
|
93 |
+
self.proj_out = nn.Conv2d(
|
94 |
+
inner_dim, in_channels, kernel_size=1, stride=1, padding=0
|
95 |
+
)
|
96 |
+
|
97 |
+
self.gradient_checkpointing = False
|
98 |
+
|
99 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
100 |
+
if hasattr(module, "gradient_checkpointing"):
|
101 |
+
module.gradient_checkpointing = value
|
102 |
+
|
103 |
+
def forward(
|
104 |
+
self,
|
105 |
+
hidden_states,
|
106 |
+
encoder_hidden_states=None,
|
107 |
+
timestep=None,
|
108 |
+
return_dict: bool = True,
|
109 |
+
):
|
110 |
+
# Input
|
111 |
+
assert (
|
112 |
+
hidden_states.dim() == 5
|
113 |
+
), f"Expected hidden_states to have ndim=5, but got ndim={hidden_states.dim()}."
|
114 |
+
video_length = hidden_states.shape[2]
|
115 |
+
hidden_states = rearrange(hidden_states, "b c f h w -> (b f) c h w")
|
116 |
+
if encoder_hidden_states.shape[0] != hidden_states.shape[0]:
|
117 |
+
encoder_hidden_states = repeat(
|
118 |
+
encoder_hidden_states, "b n c -> (b f) n c", f=video_length
|
119 |
+
)
|
120 |
+
|
121 |
+
batch, channel, height, weight = hidden_states.shape
|
122 |
+
residual = hidden_states
|
123 |
+
|
124 |
+
hidden_states = self.norm(hidden_states)
|
125 |
+
if not self.use_linear_projection:
|
126 |
+
hidden_states = self.proj_in(hidden_states)
|
127 |
+
inner_dim = hidden_states.shape[1]
|
128 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(
|
129 |
+
batch, height * weight, inner_dim
|
130 |
+
)
|
131 |
+
else:
|
132 |
+
inner_dim = hidden_states.shape[1]
|
133 |
+
hidden_states = hidden_states.permute(0, 2, 3, 1).reshape(
|
134 |
+
batch, height * weight, inner_dim
|
135 |
+
)
|
136 |
+
hidden_states = self.proj_in(hidden_states)
|
137 |
+
|
138 |
+
# Blocks
|
139 |
+
for i, block in enumerate(self.transformer_blocks):
|
140 |
+
hidden_states = block(
|
141 |
+
hidden_states,
|
142 |
+
encoder_hidden_states=encoder_hidden_states,
|
143 |
+
timestep=timestep,
|
144 |
+
video_length=video_length,
|
145 |
+
)
|
146 |
+
|
147 |
+
# Output
|
148 |
+
if not self.use_linear_projection:
|
149 |
+
hidden_states = (
|
150 |
+
hidden_states.reshape(batch, height, weight, inner_dim)
|
151 |
+
.permute(0, 3, 1, 2)
|
152 |
+
.contiguous()
|
153 |
+
)
|
154 |
+
hidden_states = self.proj_out(hidden_states)
|
155 |
+
else:
|
156 |
+
hidden_states = self.proj_out(hidden_states)
|
157 |
+
hidden_states = (
|
158 |
+
hidden_states.reshape(batch, height, weight, inner_dim)
|
159 |
+
.permute(0, 3, 1, 2)
|
160 |
+
.contiguous()
|
161 |
+
)
|
162 |
+
|
163 |
+
output = hidden_states + residual
|
164 |
+
|
165 |
+
output = rearrange(output, "(b f) c h w -> b c f h w", f=video_length)
|
166 |
+
if not return_dict:
|
167 |
+
return (output,)
|
168 |
+
|
169 |
+
return Transformer3DModelOutput(sample=output)
|
src/models/unet_2d_blocks.py
ADDED
@@ -0,0 +1,1074 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_blocks.py
|
2 |
+
from typing import Any, Dict, Optional, Tuple, Union
|
3 |
+
|
4 |
+
import numpy as np
|
5 |
+
import torch
|
6 |
+
import torch.nn.functional as F
|
7 |
+
from diffusers.models.activations import get_activation
|
8 |
+
from diffusers.models.attention_processor import Attention
|
9 |
+
from diffusers.models.dual_transformer_2d import DualTransformer2DModel
|
10 |
+
from diffusers.models.resnet import Downsample2D, ResnetBlock2D, Upsample2D
|
11 |
+
from diffusers.utils import is_torch_version, logging
|
12 |
+
from diffusers.utils.torch_utils import apply_freeu
|
13 |
+
from torch import nn
|
14 |
+
|
15 |
+
from .transformer_2d import Transformer2DModel
|
16 |
+
|
17 |
+
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
18 |
+
|
19 |
+
|
20 |
+
def get_down_block(
|
21 |
+
down_block_type: str,
|
22 |
+
num_layers: int,
|
23 |
+
in_channels: int,
|
24 |
+
out_channels: int,
|
25 |
+
temb_channels: int,
|
26 |
+
add_downsample: bool,
|
27 |
+
resnet_eps: float,
|
28 |
+
resnet_act_fn: str,
|
29 |
+
transformer_layers_per_block: int = 1,
|
30 |
+
num_attention_heads: Optional[int] = None,
|
31 |
+
resnet_groups: Optional[int] = None,
|
32 |
+
cross_attention_dim: Optional[int] = None,
|
33 |
+
downsample_padding: Optional[int] = None,
|
34 |
+
dual_cross_attention: bool = False,
|
35 |
+
use_linear_projection: bool = False,
|
36 |
+
only_cross_attention: bool = False,
|
37 |
+
upcast_attention: bool = False,
|
38 |
+
resnet_time_scale_shift: str = "default",
|
39 |
+
attention_type: str = "default",
|
40 |
+
resnet_skip_time_act: bool = False,
|
41 |
+
resnet_out_scale_factor: float = 1.0,
|
42 |
+
cross_attention_norm: Optional[str] = None,
|
43 |
+
attention_head_dim: Optional[int] = None,
|
44 |
+
downsample_type: Optional[str] = None,
|
45 |
+
dropout: float = 0.0,
|
46 |
+
):
|
47 |
+
# If attn head dim is not defined, we default it to the number of heads
|
48 |
+
if attention_head_dim is None:
|
49 |
+
logger.warn(
|
50 |
+
f"It is recommended to provide `attention_head_dim` when calling `get_down_block`. Defaulting `attention_head_dim` to {num_attention_heads}."
|
51 |
+
)
|
52 |
+
attention_head_dim = num_attention_heads
|
53 |
+
|
54 |
+
down_block_type = (
|
55 |
+
down_block_type[7:]
|
56 |
+
if down_block_type.startswith("UNetRes")
|
57 |
+
else down_block_type
|
58 |
+
)
|
59 |
+
if down_block_type == "DownBlock2D":
|
60 |
+
return DownBlock2D(
|
61 |
+
num_layers=num_layers,
|
62 |
+
in_channels=in_channels,
|
63 |
+
out_channels=out_channels,
|
64 |
+
temb_channels=temb_channels,
|
65 |
+
dropout=dropout,
|
66 |
+
add_downsample=add_downsample,
|
67 |
+
resnet_eps=resnet_eps,
|
68 |
+
resnet_act_fn=resnet_act_fn,
|
69 |
+
resnet_groups=resnet_groups,
|
70 |
+
downsample_padding=downsample_padding,
|
71 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
72 |
+
)
|
73 |
+
elif down_block_type == "CrossAttnDownBlock2D":
|
74 |
+
if cross_attention_dim is None:
|
75 |
+
raise ValueError(
|
76 |
+
"cross_attention_dim must be specified for CrossAttnDownBlock2D"
|
77 |
+
)
|
78 |
+
return CrossAttnDownBlock2D(
|
79 |
+
num_layers=num_layers,
|
80 |
+
transformer_layers_per_block=transformer_layers_per_block,
|
81 |
+
in_channels=in_channels,
|
82 |
+
out_channels=out_channels,
|
83 |
+
temb_channels=temb_channels,
|
84 |
+
dropout=dropout,
|
85 |
+
add_downsample=add_downsample,
|
86 |
+
resnet_eps=resnet_eps,
|
87 |
+
resnet_act_fn=resnet_act_fn,
|
88 |
+
resnet_groups=resnet_groups,
|
89 |
+
downsample_padding=downsample_padding,
|
90 |
+
cross_attention_dim=cross_attention_dim,
|
91 |
+
num_attention_heads=num_attention_heads,
|
92 |
+
dual_cross_attention=dual_cross_attention,
|
93 |
+
use_linear_projection=use_linear_projection,
|
94 |
+
only_cross_attention=only_cross_attention,
|
95 |
+
upcast_attention=upcast_attention,
|
96 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
97 |
+
attention_type=attention_type,
|
98 |
+
)
|
99 |
+
raise ValueError(f"{down_block_type} does not exist.")
|
100 |
+
|
101 |
+
|
102 |
+
def get_up_block(
|
103 |
+
up_block_type: str,
|
104 |
+
num_layers: int,
|
105 |
+
in_channels: int,
|
106 |
+
out_channels: int,
|
107 |
+
prev_output_channel: int,
|
108 |
+
temb_channels: int,
|
109 |
+
add_upsample: bool,
|
110 |
+
resnet_eps: float,
|
111 |
+
resnet_act_fn: str,
|
112 |
+
resolution_idx: Optional[int] = None,
|
113 |
+
transformer_layers_per_block: int = 1,
|
114 |
+
num_attention_heads: Optional[int] = None,
|
115 |
+
resnet_groups: Optional[int] = None,
|
116 |
+
cross_attention_dim: Optional[int] = None,
|
117 |
+
dual_cross_attention: bool = False,
|
118 |
+
use_linear_projection: bool = False,
|
119 |
+
only_cross_attention: bool = False,
|
120 |
+
upcast_attention: bool = False,
|
121 |
+
resnet_time_scale_shift: str = "default",
|
122 |
+
attention_type: str = "default",
|
123 |
+
resnet_skip_time_act: bool = False,
|
124 |
+
resnet_out_scale_factor: float = 1.0,
|
125 |
+
cross_attention_norm: Optional[str] = None,
|
126 |
+
attention_head_dim: Optional[int] = None,
|
127 |
+
upsample_type: Optional[str] = None,
|
128 |
+
dropout: float = 0.0,
|
129 |
+
) -> nn.Module:
|
130 |
+
# If attn head dim is not defined, we default it to the number of heads
|
131 |
+
if attention_head_dim is None:
|
132 |
+
logger.warn(
|
133 |
+
f"It is recommended to provide `attention_head_dim` when calling `get_up_block`. Defaulting `attention_head_dim` to {num_attention_heads}."
|
134 |
+
)
|
135 |
+
attention_head_dim = num_attention_heads
|
136 |
+
|
137 |
+
up_block_type = (
|
138 |
+
up_block_type[7:] if up_block_type.startswith("UNetRes") else up_block_type
|
139 |
+
)
|
140 |
+
if up_block_type == "UpBlock2D":
|
141 |
+
return UpBlock2D(
|
142 |
+
num_layers=num_layers,
|
143 |
+
in_channels=in_channels,
|
144 |
+
out_channels=out_channels,
|
145 |
+
prev_output_channel=prev_output_channel,
|
146 |
+
temb_channels=temb_channels,
|
147 |
+
resolution_idx=resolution_idx,
|
148 |
+
dropout=dropout,
|
149 |
+
add_upsample=add_upsample,
|
150 |
+
resnet_eps=resnet_eps,
|
151 |
+
resnet_act_fn=resnet_act_fn,
|
152 |
+
resnet_groups=resnet_groups,
|
153 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
154 |
+
)
|
155 |
+
elif up_block_type == "CrossAttnUpBlock2D":
|
156 |
+
if cross_attention_dim is None:
|
157 |
+
raise ValueError(
|
158 |
+
"cross_attention_dim must be specified for CrossAttnUpBlock2D"
|
159 |
+
)
|
160 |
+
return CrossAttnUpBlock2D(
|
161 |
+
num_layers=num_layers,
|
162 |
+
transformer_layers_per_block=transformer_layers_per_block,
|
163 |
+
in_channels=in_channels,
|
164 |
+
out_channels=out_channels,
|
165 |
+
prev_output_channel=prev_output_channel,
|
166 |
+
temb_channels=temb_channels,
|
167 |
+
resolution_idx=resolution_idx,
|
168 |
+
dropout=dropout,
|
169 |
+
add_upsample=add_upsample,
|
170 |
+
resnet_eps=resnet_eps,
|
171 |
+
resnet_act_fn=resnet_act_fn,
|
172 |
+
resnet_groups=resnet_groups,
|
173 |
+
cross_attention_dim=cross_attention_dim,
|
174 |
+
num_attention_heads=num_attention_heads,
|
175 |
+
dual_cross_attention=dual_cross_attention,
|
176 |
+
use_linear_projection=use_linear_projection,
|
177 |
+
only_cross_attention=only_cross_attention,
|
178 |
+
upcast_attention=upcast_attention,
|
179 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
180 |
+
attention_type=attention_type,
|
181 |
+
)
|
182 |
+
|
183 |
+
raise ValueError(f"{up_block_type} does not exist.")
|
184 |
+
|
185 |
+
|
186 |
+
class AutoencoderTinyBlock(nn.Module):
|
187 |
+
"""
|
188 |
+
Tiny Autoencoder block used in [`AutoencoderTiny`]. It is a mini residual module consisting of plain conv + ReLU
|
189 |
+
blocks.
|
190 |
+
|
191 |
+
Args:
|
192 |
+
in_channels (`int`): The number of input channels.
|
193 |
+
out_channels (`int`): The number of output channels.
|
194 |
+
act_fn (`str`):
|
195 |
+
` The activation function to use. Supported values are `"swish"`, `"mish"`, `"gelu"`, and `"relu"`.
|
196 |
+
|
197 |
+
Returns:
|
198 |
+
`torch.FloatTensor`: A tensor with the same shape as the input tensor, but with the number of channels equal to
|
199 |
+
`out_channels`.
|
200 |
+
"""
|
201 |
+
|
202 |
+
def __init__(self, in_channels: int, out_channels: int, act_fn: str):
|
203 |
+
super().__init__()
|
204 |
+
act_fn = get_activation(act_fn)
|
205 |
+
self.conv = nn.Sequential(
|
206 |
+
nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1),
|
207 |
+
act_fn,
|
208 |
+
nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1),
|
209 |
+
act_fn,
|
210 |
+
nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1),
|
211 |
+
)
|
212 |
+
self.skip = (
|
213 |
+
nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False)
|
214 |
+
if in_channels != out_channels
|
215 |
+
else nn.Identity()
|
216 |
+
)
|
217 |
+
self.fuse = nn.ReLU()
|
218 |
+
|
219 |
+
def forward(self, x: torch.FloatTensor) -> torch.FloatTensor:
|
220 |
+
return self.fuse(self.conv(x) + self.skip(x))
|
221 |
+
|
222 |
+
|
223 |
+
class UNetMidBlock2D(nn.Module):
|
224 |
+
"""
|
225 |
+
A 2D UNet mid-block [`UNetMidBlock2D`] with multiple residual blocks and optional attention blocks.
|
226 |
+
|
227 |
+
Args:
|
228 |
+
in_channels (`int`): The number of input channels.
|
229 |
+
temb_channels (`int`): The number of temporal embedding channels.
|
230 |
+
dropout (`float`, *optional*, defaults to 0.0): The dropout rate.
|
231 |
+
num_layers (`int`, *optional*, defaults to 1): The number of residual blocks.
|
232 |
+
resnet_eps (`float`, *optional*, 1e-6 ): The epsilon value for the resnet blocks.
|
233 |
+
resnet_time_scale_shift (`str`, *optional*, defaults to `default`):
|
234 |
+
The type of normalization to apply to the time embeddings. This can help to improve the performance of the
|
235 |
+
model on tasks with long-range temporal dependencies.
|
236 |
+
resnet_act_fn (`str`, *optional*, defaults to `swish`): The activation function for the resnet blocks.
|
237 |
+
resnet_groups (`int`, *optional*, defaults to 32):
|
238 |
+
The number of groups to use in the group normalization layers of the resnet blocks.
|
239 |
+
attn_groups (`Optional[int]`, *optional*, defaults to None): The number of groups for the attention blocks.
|
240 |
+
resnet_pre_norm (`bool`, *optional*, defaults to `True`):
|
241 |
+
Whether to use pre-normalization for the resnet blocks.
|
242 |
+
add_attention (`bool`, *optional*, defaults to `True`): Whether to add attention blocks.
|
243 |
+
attention_head_dim (`int`, *optional*, defaults to 1):
|
244 |
+
Dimension of a single attention head. The number of attention heads is determined based on this value and
|
245 |
+
the number of input channels.
|
246 |
+
output_scale_factor (`float`, *optional*, defaults to 1.0): The output scale factor.
|
247 |
+
|
248 |
+
Returns:
|
249 |
+
`torch.FloatTensor`: The output of the last residual block, which is a tensor of shape `(batch_size,
|
250 |
+
in_channels, height, width)`.
|
251 |
+
|
252 |
+
"""
|
253 |
+
|
254 |
+
def __init__(
|
255 |
+
self,
|
256 |
+
in_channels: int,
|
257 |
+
temb_channels: int,
|
258 |
+
dropout: float = 0.0,
|
259 |
+
num_layers: int = 1,
|
260 |
+
resnet_eps: float = 1e-6,
|
261 |
+
resnet_time_scale_shift: str = "default", # default, spatial
|
262 |
+
resnet_act_fn: str = "swish",
|
263 |
+
resnet_groups: int = 32,
|
264 |
+
attn_groups: Optional[int] = None,
|
265 |
+
resnet_pre_norm: bool = True,
|
266 |
+
add_attention: bool = True,
|
267 |
+
attention_head_dim: int = 1,
|
268 |
+
output_scale_factor: float = 1.0,
|
269 |
+
):
|
270 |
+
super().__init__()
|
271 |
+
resnet_groups = (
|
272 |
+
resnet_groups if resnet_groups is not None else min(in_channels // 4, 32)
|
273 |
+
)
|
274 |
+
self.add_attention = add_attention
|
275 |
+
|
276 |
+
if attn_groups is None:
|
277 |
+
attn_groups = (
|
278 |
+
resnet_groups if resnet_time_scale_shift == "default" else None
|
279 |
+
)
|
280 |
+
|
281 |
+
# there is always at least one resnet
|
282 |
+
resnets = [
|
283 |
+
ResnetBlock2D(
|
284 |
+
in_channels=in_channels,
|
285 |
+
out_channels=in_channels,
|
286 |
+
temb_channels=temb_channels,
|
287 |
+
eps=resnet_eps,
|
288 |
+
groups=resnet_groups,
|
289 |
+
dropout=dropout,
|
290 |
+
time_embedding_norm=resnet_time_scale_shift,
|
291 |
+
non_linearity=resnet_act_fn,
|
292 |
+
output_scale_factor=output_scale_factor,
|
293 |
+
pre_norm=resnet_pre_norm,
|
294 |
+
)
|
295 |
+
]
|
296 |
+
attentions = []
|
297 |
+
|
298 |
+
if attention_head_dim is None:
|
299 |
+
logger.warn(
|
300 |
+
f"It is not recommend to pass `attention_head_dim=None`. Defaulting `attention_head_dim` to `in_channels`: {in_channels}."
|
301 |
+
)
|
302 |
+
attention_head_dim = in_channels
|
303 |
+
|
304 |
+
for _ in range(num_layers):
|
305 |
+
if self.add_attention:
|
306 |
+
attentions.append(
|
307 |
+
Attention(
|
308 |
+
in_channels,
|
309 |
+
heads=in_channels // attention_head_dim,
|
310 |
+
dim_head=attention_head_dim,
|
311 |
+
rescale_output_factor=output_scale_factor,
|
312 |
+
eps=resnet_eps,
|
313 |
+
norm_num_groups=attn_groups,
|
314 |
+
spatial_norm_dim=temb_channels
|
315 |
+
if resnet_time_scale_shift == "spatial"
|
316 |
+
else None,
|
317 |
+
residual_connection=True,
|
318 |
+
bias=True,
|
319 |
+
upcast_softmax=True,
|
320 |
+
_from_deprecated_attn_block=True,
|
321 |
+
)
|
322 |
+
)
|
323 |
+
else:
|
324 |
+
attentions.append(None)
|
325 |
+
|
326 |
+
resnets.append(
|
327 |
+
ResnetBlock2D(
|
328 |
+
in_channels=in_channels,
|
329 |
+
out_channels=in_channels,
|
330 |
+
temb_channels=temb_channels,
|
331 |
+
eps=resnet_eps,
|
332 |
+
groups=resnet_groups,
|
333 |
+
dropout=dropout,
|
334 |
+
time_embedding_norm=resnet_time_scale_shift,
|
335 |
+
non_linearity=resnet_act_fn,
|
336 |
+
output_scale_factor=output_scale_factor,
|
337 |
+
pre_norm=resnet_pre_norm,
|
338 |
+
)
|
339 |
+
)
|
340 |
+
|
341 |
+
self.attentions = nn.ModuleList(attentions)
|
342 |
+
self.resnets = nn.ModuleList(resnets)
|
343 |
+
|
344 |
+
def forward(
|
345 |
+
self, hidden_states: torch.FloatTensor, temb: Optional[torch.FloatTensor] = None
|
346 |
+
) -> torch.FloatTensor:
|
347 |
+
hidden_states = self.resnets[0](hidden_states, temb)
|
348 |
+
for attn, resnet in zip(self.attentions, self.resnets[1:]):
|
349 |
+
if attn is not None:
|
350 |
+
hidden_states = attn(hidden_states, temb=temb)
|
351 |
+
hidden_states = resnet(hidden_states, temb)
|
352 |
+
|
353 |
+
return hidden_states
|
354 |
+
|
355 |
+
|
356 |
+
class UNetMidBlock2DCrossAttn(nn.Module):
|
357 |
+
def __init__(
|
358 |
+
self,
|
359 |
+
in_channels: int,
|
360 |
+
temb_channels: int,
|
361 |
+
dropout: float = 0.0,
|
362 |
+
num_layers: int = 1,
|
363 |
+
transformer_layers_per_block: Union[int, Tuple[int]] = 1,
|
364 |
+
resnet_eps: float = 1e-6,
|
365 |
+
resnet_time_scale_shift: str = "default",
|
366 |
+
resnet_act_fn: str = "swish",
|
367 |
+
resnet_groups: int = 32,
|
368 |
+
resnet_pre_norm: bool = True,
|
369 |
+
num_attention_heads: int = 1,
|
370 |
+
output_scale_factor: float = 1.0,
|
371 |
+
cross_attention_dim: int = 1280,
|
372 |
+
dual_cross_attention: bool = False,
|
373 |
+
use_linear_projection: bool = False,
|
374 |
+
upcast_attention: bool = False,
|
375 |
+
attention_type: str = "default",
|
376 |
+
):
|
377 |
+
super().__init__()
|
378 |
+
|
379 |
+
self.has_cross_attention = True
|
380 |
+
self.num_attention_heads = num_attention_heads
|
381 |
+
resnet_groups = (
|
382 |
+
resnet_groups if resnet_groups is not None else min(in_channels // 4, 32)
|
383 |
+
)
|
384 |
+
|
385 |
+
# support for variable transformer layers per block
|
386 |
+
if isinstance(transformer_layers_per_block, int):
|
387 |
+
transformer_layers_per_block = [transformer_layers_per_block] * num_layers
|
388 |
+
|
389 |
+
# there is always at least one resnet
|
390 |
+
resnets = [
|
391 |
+
ResnetBlock2D(
|
392 |
+
in_channels=in_channels,
|
393 |
+
out_channels=in_channels,
|
394 |
+
temb_channels=temb_channels,
|
395 |
+
eps=resnet_eps,
|
396 |
+
groups=resnet_groups,
|
397 |
+
dropout=dropout,
|
398 |
+
time_embedding_norm=resnet_time_scale_shift,
|
399 |
+
non_linearity=resnet_act_fn,
|
400 |
+
output_scale_factor=output_scale_factor,
|
401 |
+
pre_norm=resnet_pre_norm,
|
402 |
+
)
|
403 |
+
]
|
404 |
+
attentions = []
|
405 |
+
|
406 |
+
for i in range(num_layers):
|
407 |
+
if not dual_cross_attention:
|
408 |
+
attentions.append(
|
409 |
+
Transformer2DModel(
|
410 |
+
num_attention_heads,
|
411 |
+
in_channels // num_attention_heads,
|
412 |
+
in_channels=in_channels,
|
413 |
+
num_layers=transformer_layers_per_block[i],
|
414 |
+
cross_attention_dim=cross_attention_dim,
|
415 |
+
norm_num_groups=resnet_groups,
|
416 |
+
use_linear_projection=use_linear_projection,
|
417 |
+
upcast_attention=upcast_attention,
|
418 |
+
attention_type=attention_type,
|
419 |
+
)
|
420 |
+
)
|
421 |
+
else:
|
422 |
+
attentions.append(
|
423 |
+
DualTransformer2DModel(
|
424 |
+
num_attention_heads,
|
425 |
+
in_channels // num_attention_heads,
|
426 |
+
in_channels=in_channels,
|
427 |
+
num_layers=1,
|
428 |
+
cross_attention_dim=cross_attention_dim,
|
429 |
+
norm_num_groups=resnet_groups,
|
430 |
+
)
|
431 |
+
)
|
432 |
+
resnets.append(
|
433 |
+
ResnetBlock2D(
|
434 |
+
in_channels=in_channels,
|
435 |
+
out_channels=in_channels,
|
436 |
+
temb_channels=temb_channels,
|
437 |
+
eps=resnet_eps,
|
438 |
+
groups=resnet_groups,
|
439 |
+
dropout=dropout,
|
440 |
+
time_embedding_norm=resnet_time_scale_shift,
|
441 |
+
non_linearity=resnet_act_fn,
|
442 |
+
output_scale_factor=output_scale_factor,
|
443 |
+
pre_norm=resnet_pre_norm,
|
444 |
+
)
|
445 |
+
)
|
446 |
+
|
447 |
+
self.attentions = nn.ModuleList(attentions)
|
448 |
+
self.resnets = nn.ModuleList(resnets)
|
449 |
+
|
450 |
+
self.gradient_checkpointing = False
|
451 |
+
|
452 |
+
def forward(
|
453 |
+
self,
|
454 |
+
hidden_states: torch.FloatTensor,
|
455 |
+
temb: Optional[torch.FloatTensor] = None,
|
456 |
+
encoder_hidden_states: Optional[torch.FloatTensor] = None,
|
457 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
458 |
+
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
|
459 |
+
encoder_attention_mask: Optional[torch.FloatTensor] = None,
|
460 |
+
) -> torch.FloatTensor:
|
461 |
+
lora_scale = (
|
462 |
+
cross_attention_kwargs.get("scale", 1.0)
|
463 |
+
if cross_attention_kwargs is not None
|
464 |
+
else 1.0
|
465 |
+
)
|
466 |
+
hidden_states = self.resnets[0](hidden_states, temb, scale=lora_scale)
|
467 |
+
for attn, resnet in zip(self.attentions, self.resnets[1:]):
|
468 |
+
if self.training and self.gradient_checkpointing:
|
469 |
+
|
470 |
+
def create_custom_forward(module, return_dict=None):
|
471 |
+
def custom_forward(*inputs):
|
472 |
+
if return_dict is not None:
|
473 |
+
return module(*inputs, return_dict=return_dict)
|
474 |
+
else:
|
475 |
+
return module(*inputs)
|
476 |
+
|
477 |
+
return custom_forward
|
478 |
+
|
479 |
+
ckpt_kwargs: Dict[str, Any] = (
|
480 |
+
{"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
|
481 |
+
)
|
482 |
+
hidden_states, ref_feature = attn(
|
483 |
+
hidden_states,
|
484 |
+
encoder_hidden_states=encoder_hidden_states,
|
485 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
486 |
+
attention_mask=attention_mask,
|
487 |
+
encoder_attention_mask=encoder_attention_mask,
|
488 |
+
return_dict=False,
|
489 |
+
)
|
490 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
491 |
+
create_custom_forward(resnet),
|
492 |
+
hidden_states,
|
493 |
+
temb,
|
494 |
+
**ckpt_kwargs,
|
495 |
+
)
|
496 |
+
else:
|
497 |
+
hidden_states, ref_feature = attn(
|
498 |
+
hidden_states,
|
499 |
+
encoder_hidden_states=encoder_hidden_states,
|
500 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
501 |
+
attention_mask=attention_mask,
|
502 |
+
encoder_attention_mask=encoder_attention_mask,
|
503 |
+
return_dict=False,
|
504 |
+
)
|
505 |
+
hidden_states = resnet(hidden_states, temb, scale=lora_scale)
|
506 |
+
|
507 |
+
return hidden_states
|
508 |
+
|
509 |
+
|
510 |
+
class CrossAttnDownBlock2D(nn.Module):
|
511 |
+
def __init__(
|
512 |
+
self,
|
513 |
+
in_channels: int,
|
514 |
+
out_channels: int,
|
515 |
+
temb_channels: int,
|
516 |
+
dropout: float = 0.0,
|
517 |
+
num_layers: int = 1,
|
518 |
+
transformer_layers_per_block: Union[int, Tuple[int]] = 1,
|
519 |
+
resnet_eps: float = 1e-6,
|
520 |
+
resnet_time_scale_shift: str = "default",
|
521 |
+
resnet_act_fn: str = "swish",
|
522 |
+
resnet_groups: int = 32,
|
523 |
+
resnet_pre_norm: bool = True,
|
524 |
+
num_attention_heads: int = 1,
|
525 |
+
cross_attention_dim: int = 1280,
|
526 |
+
output_scale_factor: float = 1.0,
|
527 |
+
downsample_padding: int = 1,
|
528 |
+
add_downsample: bool = True,
|
529 |
+
dual_cross_attention: bool = False,
|
530 |
+
use_linear_projection: bool = False,
|
531 |
+
only_cross_attention: bool = False,
|
532 |
+
upcast_attention: bool = False,
|
533 |
+
attention_type: str = "default",
|
534 |
+
):
|
535 |
+
super().__init__()
|
536 |
+
resnets = []
|
537 |
+
attentions = []
|
538 |
+
|
539 |
+
self.has_cross_attention = True
|
540 |
+
self.num_attention_heads = num_attention_heads
|
541 |
+
if isinstance(transformer_layers_per_block, int):
|
542 |
+
transformer_layers_per_block = [transformer_layers_per_block] * num_layers
|
543 |
+
|
544 |
+
for i in range(num_layers):
|
545 |
+
in_channels = in_channels if i == 0 else out_channels
|
546 |
+
resnets.append(
|
547 |
+
ResnetBlock2D(
|
548 |
+
in_channels=in_channels,
|
549 |
+
out_channels=out_channels,
|
550 |
+
temb_channels=temb_channels,
|
551 |
+
eps=resnet_eps,
|
552 |
+
groups=resnet_groups,
|
553 |
+
dropout=dropout,
|
554 |
+
time_embedding_norm=resnet_time_scale_shift,
|
555 |
+
non_linearity=resnet_act_fn,
|
556 |
+
output_scale_factor=output_scale_factor,
|
557 |
+
pre_norm=resnet_pre_norm,
|
558 |
+
)
|
559 |
+
)
|
560 |
+
if not dual_cross_attention:
|
561 |
+
attentions.append(
|
562 |
+
Transformer2DModel(
|
563 |
+
num_attention_heads,
|
564 |
+
out_channels // num_attention_heads,
|
565 |
+
in_channels=out_channels,
|
566 |
+
num_layers=transformer_layers_per_block[i],
|
567 |
+
cross_attention_dim=cross_attention_dim,
|
568 |
+
norm_num_groups=resnet_groups,
|
569 |
+
use_linear_projection=use_linear_projection,
|
570 |
+
only_cross_attention=only_cross_attention,
|
571 |
+
upcast_attention=upcast_attention,
|
572 |
+
attention_type=attention_type,
|
573 |
+
)
|
574 |
+
)
|
575 |
+
else:
|
576 |
+
attentions.append(
|
577 |
+
DualTransformer2DModel(
|
578 |
+
num_attention_heads,
|
579 |
+
out_channels // num_attention_heads,
|
580 |
+
in_channels=out_channels,
|
581 |
+
num_layers=1,
|
582 |
+
cross_attention_dim=cross_attention_dim,
|
583 |
+
norm_num_groups=resnet_groups,
|
584 |
+
)
|
585 |
+
)
|
586 |
+
self.attentions = nn.ModuleList(attentions)
|
587 |
+
self.resnets = nn.ModuleList(resnets)
|
588 |
+
|
589 |
+
if add_downsample:
|
590 |
+
self.downsamplers = nn.ModuleList(
|
591 |
+
[
|
592 |
+
Downsample2D(
|
593 |
+
out_channels,
|
594 |
+
use_conv=True,
|
595 |
+
out_channels=out_channels,
|
596 |
+
padding=downsample_padding,
|
597 |
+
name="op",
|
598 |
+
)
|
599 |
+
]
|
600 |
+
)
|
601 |
+
else:
|
602 |
+
self.downsamplers = None
|
603 |
+
|
604 |
+
self.gradient_checkpointing = False
|
605 |
+
|
606 |
+
def forward(
|
607 |
+
self,
|
608 |
+
hidden_states: torch.FloatTensor,
|
609 |
+
temb: Optional[torch.FloatTensor] = None,
|
610 |
+
encoder_hidden_states: Optional[torch.FloatTensor] = None,
|
611 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
612 |
+
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
|
613 |
+
encoder_attention_mask: Optional[torch.FloatTensor] = None,
|
614 |
+
additional_residuals: Optional[torch.FloatTensor] = None,
|
615 |
+
) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]:
|
616 |
+
output_states = ()
|
617 |
+
|
618 |
+
lora_scale = (
|
619 |
+
cross_attention_kwargs.get("scale", 1.0)
|
620 |
+
if cross_attention_kwargs is not None
|
621 |
+
else 1.0
|
622 |
+
)
|
623 |
+
|
624 |
+
blocks = list(zip(self.resnets, self.attentions))
|
625 |
+
|
626 |
+
for i, (resnet, attn) in enumerate(blocks):
|
627 |
+
if self.training and self.gradient_checkpointing:
|
628 |
+
|
629 |
+
def create_custom_forward(module, return_dict=None):
|
630 |
+
def custom_forward(*inputs):
|
631 |
+
if return_dict is not None:
|
632 |
+
return module(*inputs, return_dict=return_dict)
|
633 |
+
else:
|
634 |
+
return module(*inputs)
|
635 |
+
|
636 |
+
return custom_forward
|
637 |
+
|
638 |
+
ckpt_kwargs: Dict[str, Any] = (
|
639 |
+
{"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
|
640 |
+
)
|
641 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
642 |
+
create_custom_forward(resnet),
|
643 |
+
hidden_states,
|
644 |
+
temb,
|
645 |
+
**ckpt_kwargs,
|
646 |
+
)
|
647 |
+
hidden_states, ref_feature = attn(
|
648 |
+
hidden_states,
|
649 |
+
encoder_hidden_states=encoder_hidden_states,
|
650 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
651 |
+
attention_mask=attention_mask,
|
652 |
+
encoder_attention_mask=encoder_attention_mask,
|
653 |
+
return_dict=False,
|
654 |
+
)
|
655 |
+
else:
|
656 |
+
hidden_states = resnet(hidden_states, temb, scale=lora_scale)
|
657 |
+
hidden_states, ref_feature = attn(
|
658 |
+
hidden_states,
|
659 |
+
encoder_hidden_states=encoder_hidden_states,
|
660 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
661 |
+
attention_mask=attention_mask,
|
662 |
+
encoder_attention_mask=encoder_attention_mask,
|
663 |
+
return_dict=False,
|
664 |
+
)
|
665 |
+
|
666 |
+
# apply additional residuals to the output of the last pair of resnet and attention blocks
|
667 |
+
if i == len(blocks) - 1 and additional_residuals is not None:
|
668 |
+
hidden_states = hidden_states + additional_residuals
|
669 |
+
|
670 |
+
output_states = output_states + (hidden_states,)
|
671 |
+
|
672 |
+
if self.downsamplers is not None:
|
673 |
+
for downsampler in self.downsamplers:
|
674 |
+
hidden_states = downsampler(hidden_states, scale=lora_scale)
|
675 |
+
|
676 |
+
output_states = output_states + (hidden_states,)
|
677 |
+
|
678 |
+
return hidden_states, output_states
|
679 |
+
|
680 |
+
|
681 |
+
class DownBlock2D(nn.Module):
|
682 |
+
def __init__(
|
683 |
+
self,
|
684 |
+
in_channels: int,
|
685 |
+
out_channels: int,
|
686 |
+
temb_channels: int,
|
687 |
+
dropout: float = 0.0,
|
688 |
+
num_layers: int = 1,
|
689 |
+
resnet_eps: float = 1e-6,
|
690 |
+
resnet_time_scale_shift: str = "default",
|
691 |
+
resnet_act_fn: str = "swish",
|
692 |
+
resnet_groups: int = 32,
|
693 |
+
resnet_pre_norm: bool = True,
|
694 |
+
output_scale_factor: float = 1.0,
|
695 |
+
add_downsample: bool = True,
|
696 |
+
downsample_padding: int = 1,
|
697 |
+
):
|
698 |
+
super().__init__()
|
699 |
+
resnets = []
|
700 |
+
|
701 |
+
for i in range(num_layers):
|
702 |
+
in_channels = in_channels if i == 0 else out_channels
|
703 |
+
resnets.append(
|
704 |
+
ResnetBlock2D(
|
705 |
+
in_channels=in_channels,
|
706 |
+
out_channels=out_channels,
|
707 |
+
temb_channels=temb_channels,
|
708 |
+
eps=resnet_eps,
|
709 |
+
groups=resnet_groups,
|
710 |
+
dropout=dropout,
|
711 |
+
time_embedding_norm=resnet_time_scale_shift,
|
712 |
+
non_linearity=resnet_act_fn,
|
713 |
+
output_scale_factor=output_scale_factor,
|
714 |
+
pre_norm=resnet_pre_norm,
|
715 |
+
)
|
716 |
+
)
|
717 |
+
|
718 |
+
self.resnets = nn.ModuleList(resnets)
|
719 |
+
|
720 |
+
if add_downsample:
|
721 |
+
self.downsamplers = nn.ModuleList(
|
722 |
+
[
|
723 |
+
Downsample2D(
|
724 |
+
out_channels,
|
725 |
+
use_conv=True,
|
726 |
+
out_channels=out_channels,
|
727 |
+
padding=downsample_padding,
|
728 |
+
name="op",
|
729 |
+
)
|
730 |
+
]
|
731 |
+
)
|
732 |
+
else:
|
733 |
+
self.downsamplers = None
|
734 |
+
|
735 |
+
self.gradient_checkpointing = False
|
736 |
+
|
737 |
+
def forward(
|
738 |
+
self,
|
739 |
+
hidden_states: torch.FloatTensor,
|
740 |
+
temb: Optional[torch.FloatTensor] = None,
|
741 |
+
scale: float = 1.0,
|
742 |
+
) -> Tuple[torch.FloatTensor, Tuple[torch.FloatTensor, ...]]:
|
743 |
+
output_states = ()
|
744 |
+
|
745 |
+
for resnet in self.resnets:
|
746 |
+
if self.training and self.gradient_checkpointing:
|
747 |
+
|
748 |
+
def create_custom_forward(module):
|
749 |
+
def custom_forward(*inputs):
|
750 |
+
return module(*inputs)
|
751 |
+
|
752 |
+
return custom_forward
|
753 |
+
|
754 |
+
if is_torch_version(">=", "1.11.0"):
|
755 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
756 |
+
create_custom_forward(resnet),
|
757 |
+
hidden_states,
|
758 |
+
temb,
|
759 |
+
use_reentrant=False,
|
760 |
+
)
|
761 |
+
else:
|
762 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
763 |
+
create_custom_forward(resnet), hidden_states, temb
|
764 |
+
)
|
765 |
+
else:
|
766 |
+
hidden_states = resnet(hidden_states, temb, scale=scale)
|
767 |
+
|
768 |
+
output_states = output_states + (hidden_states,)
|
769 |
+
|
770 |
+
if self.downsamplers is not None:
|
771 |
+
for downsampler in self.downsamplers:
|
772 |
+
hidden_states = downsampler(hidden_states, scale=scale)
|
773 |
+
|
774 |
+
output_states = output_states + (hidden_states,)
|
775 |
+
|
776 |
+
return hidden_states, output_states
|
777 |
+
|
778 |
+
|
779 |
+
class CrossAttnUpBlock2D(nn.Module):
|
780 |
+
def __init__(
|
781 |
+
self,
|
782 |
+
in_channels: int,
|
783 |
+
out_channels: int,
|
784 |
+
prev_output_channel: int,
|
785 |
+
temb_channels: int,
|
786 |
+
resolution_idx: Optional[int] = None,
|
787 |
+
dropout: float = 0.0,
|
788 |
+
num_layers: int = 1,
|
789 |
+
transformer_layers_per_block: Union[int, Tuple[int]] = 1,
|
790 |
+
resnet_eps: float = 1e-6,
|
791 |
+
resnet_time_scale_shift: str = "default",
|
792 |
+
resnet_act_fn: str = "swish",
|
793 |
+
resnet_groups: int = 32,
|
794 |
+
resnet_pre_norm: bool = True,
|
795 |
+
num_attention_heads: int = 1,
|
796 |
+
cross_attention_dim: int = 1280,
|
797 |
+
output_scale_factor: float = 1.0,
|
798 |
+
add_upsample: bool = True,
|
799 |
+
dual_cross_attention: bool = False,
|
800 |
+
use_linear_projection: bool = False,
|
801 |
+
only_cross_attention: bool = False,
|
802 |
+
upcast_attention: bool = False,
|
803 |
+
attention_type: str = "default",
|
804 |
+
):
|
805 |
+
super().__init__()
|
806 |
+
resnets = []
|
807 |
+
attentions = []
|
808 |
+
|
809 |
+
self.has_cross_attention = True
|
810 |
+
self.num_attention_heads = num_attention_heads
|
811 |
+
|
812 |
+
if isinstance(transformer_layers_per_block, int):
|
813 |
+
transformer_layers_per_block = [transformer_layers_per_block] * num_layers
|
814 |
+
|
815 |
+
for i in range(num_layers):
|
816 |
+
res_skip_channels = in_channels if (i == num_layers - 1) else out_channels
|
817 |
+
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
818 |
+
|
819 |
+
resnets.append(
|
820 |
+
ResnetBlock2D(
|
821 |
+
in_channels=resnet_in_channels + res_skip_channels,
|
822 |
+
out_channels=out_channels,
|
823 |
+
temb_channels=temb_channels,
|
824 |
+
eps=resnet_eps,
|
825 |
+
groups=resnet_groups,
|
826 |
+
dropout=dropout,
|
827 |
+
time_embedding_norm=resnet_time_scale_shift,
|
828 |
+
non_linearity=resnet_act_fn,
|
829 |
+
output_scale_factor=output_scale_factor,
|
830 |
+
pre_norm=resnet_pre_norm,
|
831 |
+
)
|
832 |
+
)
|
833 |
+
if not dual_cross_attention:
|
834 |
+
attentions.append(
|
835 |
+
Transformer2DModel(
|
836 |
+
num_attention_heads,
|
837 |
+
out_channels // num_attention_heads,
|
838 |
+
in_channels=out_channels,
|
839 |
+
num_layers=transformer_layers_per_block[i],
|
840 |
+
cross_attention_dim=cross_attention_dim,
|
841 |
+
norm_num_groups=resnet_groups,
|
842 |
+
use_linear_projection=use_linear_projection,
|
843 |
+
only_cross_attention=only_cross_attention,
|
844 |
+
upcast_attention=upcast_attention,
|
845 |
+
attention_type=attention_type,
|
846 |
+
)
|
847 |
+
)
|
848 |
+
else:
|
849 |
+
attentions.append(
|
850 |
+
DualTransformer2DModel(
|
851 |
+
num_attention_heads,
|
852 |
+
out_channels // num_attention_heads,
|
853 |
+
in_channels=out_channels,
|
854 |
+
num_layers=1,
|
855 |
+
cross_attention_dim=cross_attention_dim,
|
856 |
+
norm_num_groups=resnet_groups,
|
857 |
+
)
|
858 |
+
)
|
859 |
+
self.attentions = nn.ModuleList(attentions)
|
860 |
+
self.resnets = nn.ModuleList(resnets)
|
861 |
+
|
862 |
+
if add_upsample:
|
863 |
+
self.upsamplers = nn.ModuleList(
|
864 |
+
[Upsample2D(out_channels, use_conv=True, out_channels=out_channels)]
|
865 |
+
)
|
866 |
+
else:
|
867 |
+
self.upsamplers = None
|
868 |
+
|
869 |
+
self.gradient_checkpointing = False
|
870 |
+
self.resolution_idx = resolution_idx
|
871 |
+
|
872 |
+
def forward(
|
873 |
+
self,
|
874 |
+
hidden_states: torch.FloatTensor,
|
875 |
+
res_hidden_states_tuple: Tuple[torch.FloatTensor, ...],
|
876 |
+
temb: Optional[torch.FloatTensor] = None,
|
877 |
+
encoder_hidden_states: Optional[torch.FloatTensor] = None,
|
878 |
+
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
|
879 |
+
upsample_size: Optional[int] = None,
|
880 |
+
attention_mask: Optional[torch.FloatTensor] = None,
|
881 |
+
encoder_attention_mask: Optional[torch.FloatTensor] = None,
|
882 |
+
) -> torch.FloatTensor:
|
883 |
+
lora_scale = (
|
884 |
+
cross_attention_kwargs.get("scale", 1.0)
|
885 |
+
if cross_attention_kwargs is not None
|
886 |
+
else 1.0
|
887 |
+
)
|
888 |
+
is_freeu_enabled = (
|
889 |
+
getattr(self, "s1", None)
|
890 |
+
and getattr(self, "s2", None)
|
891 |
+
and getattr(self, "b1", None)
|
892 |
+
and getattr(self, "b2", None)
|
893 |
+
)
|
894 |
+
|
895 |
+
for resnet, attn in zip(self.resnets, self.attentions):
|
896 |
+
# pop res hidden states
|
897 |
+
res_hidden_states = res_hidden_states_tuple[-1]
|
898 |
+
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
899 |
+
|
900 |
+
# FreeU: Only operate on the first two stages
|
901 |
+
if is_freeu_enabled:
|
902 |
+
hidden_states, res_hidden_states = apply_freeu(
|
903 |
+
self.resolution_idx,
|
904 |
+
hidden_states,
|
905 |
+
res_hidden_states,
|
906 |
+
s1=self.s1,
|
907 |
+
s2=self.s2,
|
908 |
+
b1=self.b1,
|
909 |
+
b2=self.b2,
|
910 |
+
)
|
911 |
+
|
912 |
+
hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
|
913 |
+
|
914 |
+
if self.training and self.gradient_checkpointing:
|
915 |
+
|
916 |
+
def create_custom_forward(module, return_dict=None):
|
917 |
+
def custom_forward(*inputs):
|
918 |
+
if return_dict is not None:
|
919 |
+
return module(*inputs, return_dict=return_dict)
|
920 |
+
else:
|
921 |
+
return module(*inputs)
|
922 |
+
|
923 |
+
return custom_forward
|
924 |
+
|
925 |
+
ckpt_kwargs: Dict[str, Any] = (
|
926 |
+
{"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {}
|
927 |
+
)
|
928 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
929 |
+
create_custom_forward(resnet),
|
930 |
+
hidden_states,
|
931 |
+
temb,
|
932 |
+
**ckpt_kwargs,
|
933 |
+
)
|
934 |
+
hidden_states, ref_feature = attn(
|
935 |
+
hidden_states,
|
936 |
+
encoder_hidden_states=encoder_hidden_states,
|
937 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
938 |
+
attention_mask=attention_mask,
|
939 |
+
encoder_attention_mask=encoder_attention_mask,
|
940 |
+
return_dict=False,
|
941 |
+
)
|
942 |
+
else:
|
943 |
+
hidden_states = resnet(hidden_states, temb, scale=lora_scale)
|
944 |
+
hidden_states, ref_feature = attn(
|
945 |
+
hidden_states,
|
946 |
+
encoder_hidden_states=encoder_hidden_states,
|
947 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
948 |
+
attention_mask=attention_mask,
|
949 |
+
encoder_attention_mask=encoder_attention_mask,
|
950 |
+
return_dict=False,
|
951 |
+
)
|
952 |
+
|
953 |
+
if self.upsamplers is not None:
|
954 |
+
for upsampler in self.upsamplers:
|
955 |
+
hidden_states = upsampler(
|
956 |
+
hidden_states, upsample_size, scale=lora_scale
|
957 |
+
)
|
958 |
+
|
959 |
+
return hidden_states
|
960 |
+
|
961 |
+
|
962 |
+
class UpBlock2D(nn.Module):
|
963 |
+
def __init__(
|
964 |
+
self,
|
965 |
+
in_channels: int,
|
966 |
+
prev_output_channel: int,
|
967 |
+
out_channels: int,
|
968 |
+
temb_channels: int,
|
969 |
+
resolution_idx: Optional[int] = None,
|
970 |
+
dropout: float = 0.0,
|
971 |
+
num_layers: int = 1,
|
972 |
+
resnet_eps: float = 1e-6,
|
973 |
+
resnet_time_scale_shift: str = "default",
|
974 |
+
resnet_act_fn: str = "swish",
|
975 |
+
resnet_groups: int = 32,
|
976 |
+
resnet_pre_norm: bool = True,
|
977 |
+
output_scale_factor: float = 1.0,
|
978 |
+
add_upsample: bool = True,
|
979 |
+
):
|
980 |
+
super().__init__()
|
981 |
+
resnets = []
|
982 |
+
|
983 |
+
for i in range(num_layers):
|
984 |
+
res_skip_channels = in_channels if (i == num_layers - 1) else out_channels
|
985 |
+
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
986 |
+
|
987 |
+
resnets.append(
|
988 |
+
ResnetBlock2D(
|
989 |
+
in_channels=resnet_in_channels + res_skip_channels,
|
990 |
+
out_channels=out_channels,
|
991 |
+
temb_channels=temb_channels,
|
992 |
+
eps=resnet_eps,
|
993 |
+
groups=resnet_groups,
|
994 |
+
dropout=dropout,
|
995 |
+
time_embedding_norm=resnet_time_scale_shift,
|
996 |
+
non_linearity=resnet_act_fn,
|
997 |
+
output_scale_factor=output_scale_factor,
|
998 |
+
pre_norm=resnet_pre_norm,
|
999 |
+
)
|
1000 |
+
)
|
1001 |
+
|
1002 |
+
self.resnets = nn.ModuleList(resnets)
|
1003 |
+
|
1004 |
+
if add_upsample:
|
1005 |
+
self.upsamplers = nn.ModuleList(
|
1006 |
+
[Upsample2D(out_channels, use_conv=True, out_channels=out_channels)]
|
1007 |
+
)
|
1008 |
+
else:
|
1009 |
+
self.upsamplers = None
|
1010 |
+
|
1011 |
+
self.gradient_checkpointing = False
|
1012 |
+
self.resolution_idx = resolution_idx
|
1013 |
+
|
1014 |
+
def forward(
|
1015 |
+
self,
|
1016 |
+
hidden_states: torch.FloatTensor,
|
1017 |
+
res_hidden_states_tuple: Tuple[torch.FloatTensor, ...],
|
1018 |
+
temb: Optional[torch.FloatTensor] = None,
|
1019 |
+
upsample_size: Optional[int] = None,
|
1020 |
+
scale: float = 1.0,
|
1021 |
+
) -> torch.FloatTensor:
|
1022 |
+
is_freeu_enabled = (
|
1023 |
+
getattr(self, "s1", None)
|
1024 |
+
and getattr(self, "s2", None)
|
1025 |
+
and getattr(self, "b1", None)
|
1026 |
+
and getattr(self, "b2", None)
|
1027 |
+
)
|
1028 |
+
|
1029 |
+
for resnet in self.resnets:
|
1030 |
+
# pop res hidden states
|
1031 |
+
res_hidden_states = res_hidden_states_tuple[-1]
|
1032 |
+
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
1033 |
+
|
1034 |
+
# FreeU: Only operate on the first two stages
|
1035 |
+
if is_freeu_enabled:
|
1036 |
+
hidden_states, res_hidden_states = apply_freeu(
|
1037 |
+
self.resolution_idx,
|
1038 |
+
hidden_states,
|
1039 |
+
res_hidden_states,
|
1040 |
+
s1=self.s1,
|
1041 |
+
s2=self.s2,
|
1042 |
+
b1=self.b1,
|
1043 |
+
b2=self.b2,
|
1044 |
+
)
|
1045 |
+
|
1046 |
+
hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
|
1047 |
+
|
1048 |
+
if self.training and self.gradient_checkpointing:
|
1049 |
+
|
1050 |
+
def create_custom_forward(module):
|
1051 |
+
def custom_forward(*inputs):
|
1052 |
+
return module(*inputs)
|
1053 |
+
|
1054 |
+
return custom_forward
|
1055 |
+
|
1056 |
+
if is_torch_version(">=", "1.11.0"):
|
1057 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
1058 |
+
create_custom_forward(resnet),
|
1059 |
+
hidden_states,
|
1060 |
+
temb,
|
1061 |
+
use_reentrant=False,
|
1062 |
+
)
|
1063 |
+
else:
|
1064 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
1065 |
+
create_custom_forward(resnet), hidden_states, temb
|
1066 |
+
)
|
1067 |
+
else:
|
1068 |
+
hidden_states = resnet(hidden_states, temb, scale=scale)
|
1069 |
+
|
1070 |
+
if self.upsamplers is not None:
|
1071 |
+
for upsampler in self.upsamplers:
|
1072 |
+
hidden_states = upsampler(hidden_states, upsample_size, scale=scale)
|
1073 |
+
|
1074 |
+
return hidden_states
|
src/models/unet_2d_condition.py
ADDED
@@ -0,0 +1,1308 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_condition.py
|
2 |
+
from dataclasses import dataclass
|
3 |
+
from typing import Any, Dict, List, Optional, Tuple, Union
|
4 |
+
|
5 |
+
import torch
|
6 |
+
import torch.nn as nn
|
7 |
+
import torch.utils.checkpoint
|
8 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
9 |
+
from diffusers.loaders import UNet2DConditionLoadersMixin
|
10 |
+
from diffusers.models.activations import get_activation
|
11 |
+
from diffusers.models.attention_processor import (
|
12 |
+
ADDED_KV_ATTENTION_PROCESSORS,
|
13 |
+
CROSS_ATTENTION_PROCESSORS,
|
14 |
+
AttentionProcessor,
|
15 |
+
AttnAddedKVProcessor,
|
16 |
+
AttnProcessor,
|
17 |
+
)
|
18 |
+
from diffusers.models.embeddings import (
|
19 |
+
GaussianFourierProjection,
|
20 |
+
ImageHintTimeEmbedding,
|
21 |
+
ImageProjection,
|
22 |
+
ImageTimeEmbedding,
|
23 |
+
PositionNet,
|
24 |
+
TextImageProjection,
|
25 |
+
TextImageTimeEmbedding,
|
26 |
+
TextTimeEmbedding,
|
27 |
+
TimestepEmbedding,
|
28 |
+
Timesteps,
|
29 |
+
)
|
30 |
+
from diffusers.models.modeling_utils import ModelMixin
|
31 |
+
from diffusers.utils import (
|
32 |
+
USE_PEFT_BACKEND,
|
33 |
+
BaseOutput,
|
34 |
+
deprecate,
|
35 |
+
logging,
|
36 |
+
scale_lora_layers,
|
37 |
+
unscale_lora_layers,
|
38 |
+
)
|
39 |
+
|
40 |
+
from .unet_2d_blocks import (
|
41 |
+
UNetMidBlock2D,
|
42 |
+
UNetMidBlock2DCrossAttn,
|
43 |
+
get_down_block,
|
44 |
+
get_up_block,
|
45 |
+
)
|
46 |
+
|
47 |
+
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
48 |
+
|
49 |
+
|
50 |
+
@dataclass
|
51 |
+
class UNet2DConditionOutput(BaseOutput):
|
52 |
+
"""
|
53 |
+
The output of [`UNet2DConditionModel`].
|
54 |
+
|
55 |
+
Args:
|
56 |
+
sample (`torch.FloatTensor` of shape `(batch_size, num_channels, height, width)`):
|
57 |
+
The hidden states output conditioned on `encoder_hidden_states` input. Output of last layer of model.
|
58 |
+
"""
|
59 |
+
|
60 |
+
sample: torch.FloatTensor = None
|
61 |
+
ref_features: Tuple[torch.FloatTensor] = None
|
62 |
+
|
63 |
+
|
64 |
+
class UNet2DConditionModel(ModelMixin, ConfigMixin, UNet2DConditionLoadersMixin):
|
65 |
+
r"""
|
66 |
+
A conditional 2D UNet model that takes a noisy sample, conditional state, and a timestep and returns a sample
|
67 |
+
shaped output.
|
68 |
+
|
69 |
+
This model inherits from [`ModelMixin`]. Check the superclass documentation for it's generic methods implemented
|
70 |
+
for all models (such as downloading or saving).
|
71 |
+
|
72 |
+
Parameters:
|
73 |
+
sample_size (`int` or `Tuple[int, int]`, *optional*, defaults to `None`):
|
74 |
+
Height and width of input/output sample.
|
75 |
+
in_channels (`int`, *optional*, defaults to 4): Number of channels in the input sample.
|
76 |
+
out_channels (`int`, *optional*, defaults to 4): Number of channels in the output.
|
77 |
+
center_input_sample (`bool`, *optional*, defaults to `False`): Whether to center the input sample.
|
78 |
+
flip_sin_to_cos (`bool`, *optional*, defaults to `False`):
|
79 |
+
Whether to flip the sin to cos in the time embedding.
|
80 |
+
freq_shift (`int`, *optional*, defaults to 0): The frequency shift to apply to the time embedding.
|
81 |
+
down_block_types (`Tuple[str]`, *optional*, defaults to `("CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "CrossAttnDownBlock2D", "DownBlock2D")`):
|
82 |
+
The tuple of downsample blocks to use.
|
83 |
+
mid_block_type (`str`, *optional*, defaults to `"UNetMidBlock2DCrossAttn"`):
|
84 |
+
Block type for middle of UNet, it can be one of `UNetMidBlock2DCrossAttn`, `UNetMidBlock2D`, or
|
85 |
+
`UNetMidBlock2DSimpleCrossAttn`. If `None`, the mid block layer is skipped.
|
86 |
+
up_block_types (`Tuple[str]`, *optional*, defaults to `("UpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D", "CrossAttnUpBlock2D")`):
|
87 |
+
The tuple of upsample blocks to use.
|
88 |
+
only_cross_attention(`bool` or `Tuple[bool]`, *optional*, default to `False`):
|
89 |
+
Whether to include self-attention in the basic transformer blocks, see
|
90 |
+
[`~models.attention.BasicTransformerBlock`].
|
91 |
+
block_out_channels (`Tuple[int]`, *optional*, defaults to `(320, 640, 1280, 1280)`):
|
92 |
+
The tuple of output channels for each block.
|
93 |
+
layers_per_block (`int`, *optional*, defaults to 2): The number of layers per block.
|
94 |
+
downsample_padding (`int`, *optional*, defaults to 1): The padding to use for the downsampling convolution.
|
95 |
+
mid_block_scale_factor (`float`, *optional*, defaults to 1.0): The scale factor to use for the mid block.
|
96 |
+
dropout (`float`, *optional*, defaults to 0.0): The dropout probability to use.
|
97 |
+
act_fn (`str`, *optional*, defaults to `"silu"`): The activation function to use.
|
98 |
+
norm_num_groups (`int`, *optional*, defaults to 32): The number of groups to use for the normalization.
|
99 |
+
If `None`, normalization and activation layers is skipped in post-processing.
|
100 |
+
norm_eps (`float`, *optional*, defaults to 1e-5): The epsilon to use for the normalization.
|
101 |
+
cross_attention_dim (`int` or `Tuple[int]`, *optional*, defaults to 1280):
|
102 |
+
The dimension of the cross attention features.
|
103 |
+
transformer_layers_per_block (`int`, `Tuple[int]`, or `Tuple[Tuple]` , *optional*, defaults to 1):
|
104 |
+
The number of transformer blocks of type [`~models.attention.BasicTransformerBlock`]. Only relevant for
|
105 |
+
[`~models.unet_2d_blocks.CrossAttnDownBlock2D`], [`~models.unet_2d_blocks.CrossAttnUpBlock2D`],
|
106 |
+
[`~models.unet_2d_blocks.UNetMidBlock2DCrossAttn`].
|
107 |
+
reverse_transformer_layers_per_block : (`Tuple[Tuple]`, *optional*, defaults to None):
|
108 |
+
The number of transformer blocks of type [`~models.attention.BasicTransformerBlock`], in the upsampling
|
109 |
+
blocks of the U-Net. Only relevant if `transformer_layers_per_block` is of type `Tuple[Tuple]` and for
|
110 |
+
[`~models.unet_2d_blocks.CrossAttnDownBlock2D`], [`~models.unet_2d_blocks.CrossAttnUpBlock2D`],
|
111 |
+
[`~models.unet_2d_blocks.UNetMidBlock2DCrossAttn`].
|
112 |
+
encoder_hid_dim (`int`, *optional*, defaults to None):
|
113 |
+
If `encoder_hid_dim_type` is defined, `encoder_hidden_states` will be projected from `encoder_hid_dim`
|
114 |
+
dimension to `cross_attention_dim`.
|
115 |
+
encoder_hid_dim_type (`str`, *optional*, defaults to `None`):
|
116 |
+
If given, the `encoder_hidden_states` and potentially other embeddings are down-projected to text
|
117 |
+
embeddings of dimension `cross_attention` according to `encoder_hid_dim_type`.
|
118 |
+
attention_head_dim (`int`, *optional*, defaults to 8): The dimension of the attention heads.
|
119 |
+
num_attention_heads (`int`, *optional*):
|
120 |
+
The number of attention heads. If not defined, defaults to `attention_head_dim`
|
121 |
+
resnet_time_scale_shift (`str`, *optional*, defaults to `"default"`): Time scale shift config
|
122 |
+
for ResNet blocks (see [`~models.resnet.ResnetBlock2D`]). Choose from `default` or `scale_shift`.
|
123 |
+
class_embed_type (`str`, *optional*, defaults to `None`):
|
124 |
+
The type of class embedding to use which is ultimately summed with the time embeddings. Choose from `None`,
|
125 |
+
`"timestep"`, `"identity"`, `"projection"`, or `"simple_projection"`.
|
126 |
+
addition_embed_type (`str`, *optional*, defaults to `None`):
|
127 |
+
Configures an optional embedding which will be summed with the time embeddings. Choose from `None` or
|
128 |
+
"text". "text" will use the `TextTimeEmbedding` layer.
|
129 |
+
addition_time_embed_dim: (`int`, *optional*, defaults to `None`):
|
130 |
+
Dimension for the timestep embeddings.
|
131 |
+
num_class_embeds (`int`, *optional*, defaults to `None`):
|
132 |
+
Input dimension of the learnable embedding matrix to be projected to `time_embed_dim`, when performing
|
133 |
+
class conditioning with `class_embed_type` equal to `None`.
|
134 |
+
time_embedding_type (`str`, *optional*, defaults to `positional`):
|
135 |
+
The type of position embedding to use for timesteps. Choose from `positional` or `fourier`.
|
136 |
+
time_embedding_dim (`int`, *optional*, defaults to `None`):
|
137 |
+
An optional override for the dimension of the projected time embedding.
|
138 |
+
time_embedding_act_fn (`str`, *optional*, defaults to `None`):
|
139 |
+
Optional activation function to use only once on the time embeddings before they are passed to the rest of
|
140 |
+
the UNet. Choose from `silu`, `mish`, `gelu`, and `swish`.
|
141 |
+
timestep_post_act (`str`, *optional*, defaults to `None`):
|
142 |
+
The second activation function to use in timestep embedding. Choose from `silu`, `mish` and `gelu`.
|
143 |
+
time_cond_proj_dim (`int`, *optional*, defaults to `None`):
|
144 |
+
The dimension of `cond_proj` layer in the timestep embedding.
|
145 |
+
conv_in_kernel (`int`, *optional*, default to `3`): The kernel size of `conv_in` layer. conv_out_kernel (`int`,
|
146 |
+
*optional*, default to `3`): The kernel size of `conv_out` layer. projection_class_embeddings_input_dim (`int`,
|
147 |
+
*optional*): The dimension of the `class_labels` input when
|
148 |
+
`class_embed_type="projection"`. Required when `class_embed_type="projection"`.
|
149 |
+
class_embeddings_concat (`bool`, *optional*, defaults to `False`): Whether to concatenate the time
|
150 |
+
embeddings with the class embeddings.
|
151 |
+
mid_block_only_cross_attention (`bool`, *optional*, defaults to `None`):
|
152 |
+
Whether to use cross attention with the mid block when using the `UNetMidBlock2DSimpleCrossAttn`. If
|
153 |
+
`only_cross_attention` is given as a single boolean and `mid_block_only_cross_attention` is `None`, the
|
154 |
+
`only_cross_attention` value is used as the value for `mid_block_only_cross_attention`. Default to `False`
|
155 |
+
otherwise.
|
156 |
+
"""
|
157 |
+
|
158 |
+
_supports_gradient_checkpointing = True
|
159 |
+
|
160 |
+
@register_to_config
|
161 |
+
def __init__(
|
162 |
+
self,
|
163 |
+
sample_size: Optional[int] = None,
|
164 |
+
in_channels: int = 4,
|
165 |
+
out_channels: int = 4,
|
166 |
+
center_input_sample: bool = False,
|
167 |
+
flip_sin_to_cos: bool = True,
|
168 |
+
freq_shift: int = 0,
|
169 |
+
down_block_types: Tuple[str] = (
|
170 |
+
"CrossAttnDownBlock2D",
|
171 |
+
"CrossAttnDownBlock2D",
|
172 |
+
"CrossAttnDownBlock2D",
|
173 |
+
"DownBlock2D",
|
174 |
+
),
|
175 |
+
mid_block_type: Optional[str] = "UNetMidBlock2DCrossAttn",
|
176 |
+
up_block_types: Tuple[str] = (
|
177 |
+
"UpBlock2D",
|
178 |
+
"CrossAttnUpBlock2D",
|
179 |
+
"CrossAttnUpBlock2D",
|
180 |
+
"CrossAttnUpBlock2D",
|
181 |
+
),
|
182 |
+
only_cross_attention: Union[bool, Tuple[bool]] = False,
|
183 |
+
block_out_channels: Tuple[int] = (320, 640, 1280, 1280),
|
184 |
+
layers_per_block: Union[int, Tuple[int]] = 2,
|
185 |
+
downsample_padding: int = 1,
|
186 |
+
mid_block_scale_factor: float = 1,
|
187 |
+
dropout: float = 0.0,
|
188 |
+
act_fn: str = "silu",
|
189 |
+
norm_num_groups: Optional[int] = 32,
|
190 |
+
norm_eps: float = 1e-5,
|
191 |
+
cross_attention_dim: Union[int, Tuple[int]] = 1280,
|
192 |
+
transformer_layers_per_block: Union[int, Tuple[int], Tuple[Tuple]] = 1,
|
193 |
+
reverse_transformer_layers_per_block: Optional[Tuple[Tuple[int]]] = None,
|
194 |
+
encoder_hid_dim: Optional[int] = None,
|
195 |
+
encoder_hid_dim_type: Optional[str] = None,
|
196 |
+
attention_head_dim: Union[int, Tuple[int]] = 8,
|
197 |
+
num_attention_heads: Optional[Union[int, Tuple[int]]] = None,
|
198 |
+
dual_cross_attention: bool = False,
|
199 |
+
use_linear_projection: bool = False,
|
200 |
+
class_embed_type: Optional[str] = None,
|
201 |
+
addition_embed_type: Optional[str] = None,
|
202 |
+
addition_time_embed_dim: Optional[int] = None,
|
203 |
+
num_class_embeds: Optional[int] = None,
|
204 |
+
upcast_attention: bool = False,
|
205 |
+
resnet_time_scale_shift: str = "default",
|
206 |
+
resnet_skip_time_act: bool = False,
|
207 |
+
resnet_out_scale_factor: int = 1.0,
|
208 |
+
time_embedding_type: str = "positional",
|
209 |
+
time_embedding_dim: Optional[int] = None,
|
210 |
+
time_embedding_act_fn: Optional[str] = None,
|
211 |
+
timestep_post_act: Optional[str] = None,
|
212 |
+
time_cond_proj_dim: Optional[int] = None,
|
213 |
+
conv_in_kernel: int = 3,
|
214 |
+
conv_out_kernel: int = 3,
|
215 |
+
projection_class_embeddings_input_dim: Optional[int] = None,
|
216 |
+
attention_type: str = "default",
|
217 |
+
class_embeddings_concat: bool = False,
|
218 |
+
mid_block_only_cross_attention: Optional[bool] = None,
|
219 |
+
cross_attention_norm: Optional[str] = None,
|
220 |
+
addition_embed_type_num_heads=64,
|
221 |
+
):
|
222 |
+
super().__init__()
|
223 |
+
|
224 |
+
self.sample_size = sample_size
|
225 |
+
|
226 |
+
if num_attention_heads is not None:
|
227 |
+
raise ValueError(
|
228 |
+
"At the moment it is not possible to define the number of attention heads via `num_attention_heads` because of a naming issue as described in https://github.com/huggingface/diffusers/issues/2011#issuecomment-1547958131. Passing `num_attention_heads` will only be supported in diffusers v0.19."
|
229 |
+
)
|
230 |
+
|
231 |
+
# If `num_attention_heads` is not defined (which is the case for most models)
|
232 |
+
# it will default to `attention_head_dim`. This looks weird upon first reading it and it is.
|
233 |
+
# The reason for this behavior is to correct for incorrectly named variables that were introduced
|
234 |
+
# when this library was created. The incorrect naming was only discovered much later in https://github.com/huggingface/diffusers/issues/2011#issuecomment-1547958131
|
235 |
+
# Changing `attention_head_dim` to `num_attention_heads` for 40,000+ configurations is too backwards breaking
|
236 |
+
# which is why we correct for the naming here.
|
237 |
+
num_attention_heads = num_attention_heads or attention_head_dim
|
238 |
+
|
239 |
+
# Check inputs
|
240 |
+
if len(down_block_types) != len(up_block_types):
|
241 |
+
raise ValueError(
|
242 |
+
f"Must provide the same number of `down_block_types` as `up_block_types`. `down_block_types`: {down_block_types}. `up_block_types`: {up_block_types}."
|
243 |
+
)
|
244 |
+
|
245 |
+
if len(block_out_channels) != len(down_block_types):
|
246 |
+
raise ValueError(
|
247 |
+
f"Must provide the same number of `block_out_channels` as `down_block_types`. `block_out_channels`: {block_out_channels}. `down_block_types`: {down_block_types}."
|
248 |
+
)
|
249 |
+
|
250 |
+
if not isinstance(only_cross_attention, bool) and len(
|
251 |
+
only_cross_attention
|
252 |
+
) != len(down_block_types):
|
253 |
+
raise ValueError(
|
254 |
+
f"Must provide the same number of `only_cross_attention` as `down_block_types`. `only_cross_attention`: {only_cross_attention}. `down_block_types`: {down_block_types}."
|
255 |
+
)
|
256 |
+
|
257 |
+
if not isinstance(num_attention_heads, int) and len(num_attention_heads) != len(
|
258 |
+
down_block_types
|
259 |
+
):
|
260 |
+
raise ValueError(
|
261 |
+
f"Must provide the same number of `num_attention_heads` as `down_block_types`. `num_attention_heads`: {num_attention_heads}. `down_block_types`: {down_block_types}."
|
262 |
+
)
|
263 |
+
|
264 |
+
if not isinstance(attention_head_dim, int) and len(attention_head_dim) != len(
|
265 |
+
down_block_types
|
266 |
+
):
|
267 |
+
raise ValueError(
|
268 |
+
f"Must provide the same number of `attention_head_dim` as `down_block_types`. `attention_head_dim`: {attention_head_dim}. `down_block_types`: {down_block_types}."
|
269 |
+
)
|
270 |
+
|
271 |
+
if isinstance(cross_attention_dim, list) and len(cross_attention_dim) != len(
|
272 |
+
down_block_types
|
273 |
+
):
|
274 |
+
raise ValueError(
|
275 |
+
f"Must provide the same number of `cross_attention_dim` as `down_block_types`. `cross_attention_dim`: {cross_attention_dim}. `down_block_types`: {down_block_types}."
|
276 |
+
)
|
277 |
+
|
278 |
+
if not isinstance(layers_per_block, int) and len(layers_per_block) != len(
|
279 |
+
down_block_types
|
280 |
+
):
|
281 |
+
raise ValueError(
|
282 |
+
f"Must provide the same number of `layers_per_block` as `down_block_types`. `layers_per_block`: {layers_per_block}. `down_block_types`: {down_block_types}."
|
283 |
+
)
|
284 |
+
if (
|
285 |
+
isinstance(transformer_layers_per_block, list)
|
286 |
+
and reverse_transformer_layers_per_block is None
|
287 |
+
):
|
288 |
+
for layer_number_per_block in transformer_layers_per_block:
|
289 |
+
if isinstance(layer_number_per_block, list):
|
290 |
+
raise ValueError(
|
291 |
+
"Must provide 'reverse_transformer_layers_per_block` if using asymmetrical UNet."
|
292 |
+
)
|
293 |
+
|
294 |
+
# input
|
295 |
+
conv_in_padding = (conv_in_kernel - 1) // 2
|
296 |
+
self.conv_in = nn.Conv2d(
|
297 |
+
in_channels,
|
298 |
+
block_out_channels[0],
|
299 |
+
kernel_size=conv_in_kernel,
|
300 |
+
padding=conv_in_padding,
|
301 |
+
)
|
302 |
+
|
303 |
+
# time
|
304 |
+
if time_embedding_type == "fourier":
|
305 |
+
time_embed_dim = time_embedding_dim or block_out_channels[0] * 2
|
306 |
+
if time_embed_dim % 2 != 0:
|
307 |
+
raise ValueError(
|
308 |
+
f"`time_embed_dim` should be divisible by 2, but is {time_embed_dim}."
|
309 |
+
)
|
310 |
+
self.time_proj = GaussianFourierProjection(
|
311 |
+
time_embed_dim // 2,
|
312 |
+
set_W_to_weight=False,
|
313 |
+
log=False,
|
314 |
+
flip_sin_to_cos=flip_sin_to_cos,
|
315 |
+
)
|
316 |
+
timestep_input_dim = time_embed_dim
|
317 |
+
elif time_embedding_type == "positional":
|
318 |
+
time_embed_dim = time_embedding_dim or block_out_channels[0] * 4
|
319 |
+
|
320 |
+
self.time_proj = Timesteps(
|
321 |
+
block_out_channels[0], flip_sin_to_cos, freq_shift
|
322 |
+
)
|
323 |
+
timestep_input_dim = block_out_channels[0]
|
324 |
+
else:
|
325 |
+
raise ValueError(
|
326 |
+
f"{time_embedding_type} does not exist. Please make sure to use one of `fourier` or `positional`."
|
327 |
+
)
|
328 |
+
|
329 |
+
self.time_embedding = TimestepEmbedding(
|
330 |
+
timestep_input_dim,
|
331 |
+
time_embed_dim,
|
332 |
+
act_fn=act_fn,
|
333 |
+
post_act_fn=timestep_post_act,
|
334 |
+
cond_proj_dim=time_cond_proj_dim,
|
335 |
+
)
|
336 |
+
|
337 |
+
if encoder_hid_dim_type is None and encoder_hid_dim is not None:
|
338 |
+
encoder_hid_dim_type = "text_proj"
|
339 |
+
self.register_to_config(encoder_hid_dim_type=encoder_hid_dim_type)
|
340 |
+
logger.info(
|
341 |
+
"encoder_hid_dim_type defaults to 'text_proj' as `encoder_hid_dim` is defined."
|
342 |
+
)
|
343 |
+
|
344 |
+
if encoder_hid_dim is None and encoder_hid_dim_type is not None:
|
345 |
+
raise ValueError(
|
346 |
+
f"`encoder_hid_dim` has to be defined when `encoder_hid_dim_type` is set to {encoder_hid_dim_type}."
|
347 |
+
)
|
348 |
+
|
349 |
+
if encoder_hid_dim_type == "text_proj":
|
350 |
+
self.encoder_hid_proj = nn.Linear(encoder_hid_dim, cross_attention_dim)
|
351 |
+
elif encoder_hid_dim_type == "text_image_proj":
|
352 |
+
# image_embed_dim DOESN'T have to be `cross_attention_dim`. To not clutter the __init__ too much
|
353 |
+
# they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
|
354 |
+
# case when `addition_embed_type == "text_image_proj"` (Kadinsky 2.1)`
|
355 |
+
self.encoder_hid_proj = TextImageProjection(
|
356 |
+
text_embed_dim=encoder_hid_dim,
|
357 |
+
image_embed_dim=cross_attention_dim,
|
358 |
+
cross_attention_dim=cross_attention_dim,
|
359 |
+
)
|
360 |
+
elif encoder_hid_dim_type == "image_proj":
|
361 |
+
# Kandinsky 2.2
|
362 |
+
self.encoder_hid_proj = ImageProjection(
|
363 |
+
image_embed_dim=encoder_hid_dim,
|
364 |
+
cross_attention_dim=cross_attention_dim,
|
365 |
+
)
|
366 |
+
elif encoder_hid_dim_type is not None:
|
367 |
+
raise ValueError(
|
368 |
+
f"encoder_hid_dim_type: {encoder_hid_dim_type} must be None, 'text_proj' or 'text_image_proj'."
|
369 |
+
)
|
370 |
+
else:
|
371 |
+
self.encoder_hid_proj = None
|
372 |
+
|
373 |
+
# class embedding
|
374 |
+
if class_embed_type is None and num_class_embeds is not None:
|
375 |
+
self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
|
376 |
+
elif class_embed_type == "timestep":
|
377 |
+
self.class_embedding = TimestepEmbedding(
|
378 |
+
timestep_input_dim, time_embed_dim, act_fn=act_fn
|
379 |
+
)
|
380 |
+
elif class_embed_type == "identity":
|
381 |
+
self.class_embedding = nn.Identity(time_embed_dim, time_embed_dim)
|
382 |
+
elif class_embed_type == "projection":
|
383 |
+
if projection_class_embeddings_input_dim is None:
|
384 |
+
raise ValueError(
|
385 |
+
"`class_embed_type`: 'projection' requires `projection_class_embeddings_input_dim` be set"
|
386 |
+
)
|
387 |
+
# The projection `class_embed_type` is the same as the timestep `class_embed_type` except
|
388 |
+
# 1. the `class_labels` inputs are not first converted to sinusoidal embeddings
|
389 |
+
# 2. it projects from an arbitrary input dimension.
|
390 |
+
#
|
391 |
+
# Note that `TimestepEmbedding` is quite general, being mainly linear layers and activations.
|
392 |
+
# When used for embedding actual timesteps, the timesteps are first converted to sinusoidal embeddings.
|
393 |
+
# As a result, `TimestepEmbedding` can be passed arbitrary vectors.
|
394 |
+
self.class_embedding = TimestepEmbedding(
|
395 |
+
projection_class_embeddings_input_dim, time_embed_dim
|
396 |
+
)
|
397 |
+
elif class_embed_type == "simple_projection":
|
398 |
+
if projection_class_embeddings_input_dim is None:
|
399 |
+
raise ValueError(
|
400 |
+
"`class_embed_type`: 'simple_projection' requires `projection_class_embeddings_input_dim` be set"
|
401 |
+
)
|
402 |
+
self.class_embedding = nn.Linear(
|
403 |
+
projection_class_embeddings_input_dim, time_embed_dim
|
404 |
+
)
|
405 |
+
else:
|
406 |
+
self.class_embedding = None
|
407 |
+
|
408 |
+
if addition_embed_type == "text":
|
409 |
+
if encoder_hid_dim is not None:
|
410 |
+
text_time_embedding_from_dim = encoder_hid_dim
|
411 |
+
else:
|
412 |
+
text_time_embedding_from_dim = cross_attention_dim
|
413 |
+
|
414 |
+
self.add_embedding = TextTimeEmbedding(
|
415 |
+
text_time_embedding_from_dim,
|
416 |
+
time_embed_dim,
|
417 |
+
num_heads=addition_embed_type_num_heads,
|
418 |
+
)
|
419 |
+
elif addition_embed_type == "text_image":
|
420 |
+
# text_embed_dim and image_embed_dim DON'T have to be `cross_attention_dim`. To not clutter the __init__ too much
|
421 |
+
# they are set to `cross_attention_dim` here as this is exactly the required dimension for the currently only use
|
422 |
+
# case when `addition_embed_type == "text_image"` (Kadinsky 2.1)`
|
423 |
+
self.add_embedding = TextImageTimeEmbedding(
|
424 |
+
text_embed_dim=cross_attention_dim,
|
425 |
+
image_embed_dim=cross_attention_dim,
|
426 |
+
time_embed_dim=time_embed_dim,
|
427 |
+
)
|
428 |
+
elif addition_embed_type == "text_time":
|
429 |
+
self.add_time_proj = Timesteps(
|
430 |
+
addition_time_embed_dim, flip_sin_to_cos, freq_shift
|
431 |
+
)
|
432 |
+
self.add_embedding = TimestepEmbedding(
|
433 |
+
projection_class_embeddings_input_dim, time_embed_dim
|
434 |
+
)
|
435 |
+
elif addition_embed_type == "image":
|
436 |
+
# Kandinsky 2.2
|
437 |
+
self.add_embedding = ImageTimeEmbedding(
|
438 |
+
image_embed_dim=encoder_hid_dim, time_embed_dim=time_embed_dim
|
439 |
+
)
|
440 |
+
elif addition_embed_type == "image_hint":
|
441 |
+
# Kandinsky 2.2 ControlNet
|
442 |
+
self.add_embedding = ImageHintTimeEmbedding(
|
443 |
+
image_embed_dim=encoder_hid_dim, time_embed_dim=time_embed_dim
|
444 |
+
)
|
445 |
+
elif addition_embed_type is not None:
|
446 |
+
raise ValueError(
|
447 |
+
f"addition_embed_type: {addition_embed_type} must be None, 'text' or 'text_image'."
|
448 |
+
)
|
449 |
+
|
450 |
+
if time_embedding_act_fn is None:
|
451 |
+
self.time_embed_act = None
|
452 |
+
else:
|
453 |
+
self.time_embed_act = get_activation(time_embedding_act_fn)
|
454 |
+
|
455 |
+
self.down_blocks = nn.ModuleList([])
|
456 |
+
self.up_blocks = nn.ModuleList([])
|
457 |
+
|
458 |
+
if isinstance(only_cross_attention, bool):
|
459 |
+
if mid_block_only_cross_attention is None:
|
460 |
+
mid_block_only_cross_attention = only_cross_attention
|
461 |
+
|
462 |
+
only_cross_attention = [only_cross_attention] * len(down_block_types)
|
463 |
+
|
464 |
+
if mid_block_only_cross_attention is None:
|
465 |
+
mid_block_only_cross_attention = False
|
466 |
+
|
467 |
+
if isinstance(num_attention_heads, int):
|
468 |
+
num_attention_heads = (num_attention_heads,) * len(down_block_types)
|
469 |
+
|
470 |
+
if isinstance(attention_head_dim, int):
|
471 |
+
attention_head_dim = (attention_head_dim,) * len(down_block_types)
|
472 |
+
|
473 |
+
if isinstance(cross_attention_dim, int):
|
474 |
+
cross_attention_dim = (cross_attention_dim,) * len(down_block_types)
|
475 |
+
|
476 |
+
if isinstance(layers_per_block, int):
|
477 |
+
layers_per_block = [layers_per_block] * len(down_block_types)
|
478 |
+
|
479 |
+
if isinstance(transformer_layers_per_block, int):
|
480 |
+
transformer_layers_per_block = [transformer_layers_per_block] * len(
|
481 |
+
down_block_types
|
482 |
+
)
|
483 |
+
|
484 |
+
if class_embeddings_concat:
|
485 |
+
# The time embeddings are concatenated with the class embeddings. The dimension of the
|
486 |
+
# time embeddings passed to the down, middle, and up blocks is twice the dimension of the
|
487 |
+
# regular time embeddings
|
488 |
+
blocks_time_embed_dim = time_embed_dim * 2
|
489 |
+
else:
|
490 |
+
blocks_time_embed_dim = time_embed_dim
|
491 |
+
|
492 |
+
# down
|
493 |
+
output_channel = block_out_channels[0]
|
494 |
+
for i, down_block_type in enumerate(down_block_types):
|
495 |
+
input_channel = output_channel
|
496 |
+
output_channel = block_out_channels[i]
|
497 |
+
is_final_block = i == len(block_out_channels) - 1
|
498 |
+
|
499 |
+
down_block = get_down_block(
|
500 |
+
down_block_type,
|
501 |
+
num_layers=layers_per_block[i],
|
502 |
+
transformer_layers_per_block=transformer_layers_per_block[i],
|
503 |
+
in_channels=input_channel,
|
504 |
+
out_channels=output_channel,
|
505 |
+
temb_channels=blocks_time_embed_dim,
|
506 |
+
add_downsample=not is_final_block,
|
507 |
+
resnet_eps=norm_eps,
|
508 |
+
resnet_act_fn=act_fn,
|
509 |
+
resnet_groups=norm_num_groups,
|
510 |
+
cross_attention_dim=cross_attention_dim[i],
|
511 |
+
num_attention_heads=num_attention_heads[i],
|
512 |
+
downsample_padding=downsample_padding,
|
513 |
+
dual_cross_attention=dual_cross_attention,
|
514 |
+
use_linear_projection=use_linear_projection,
|
515 |
+
only_cross_attention=only_cross_attention[i],
|
516 |
+
upcast_attention=upcast_attention,
|
517 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
518 |
+
attention_type=attention_type,
|
519 |
+
resnet_skip_time_act=resnet_skip_time_act,
|
520 |
+
resnet_out_scale_factor=resnet_out_scale_factor,
|
521 |
+
cross_attention_norm=cross_attention_norm,
|
522 |
+
attention_head_dim=attention_head_dim[i]
|
523 |
+
if attention_head_dim[i] is not None
|
524 |
+
else output_channel,
|
525 |
+
dropout=dropout,
|
526 |
+
)
|
527 |
+
self.down_blocks.append(down_block)
|
528 |
+
|
529 |
+
# mid
|
530 |
+
if mid_block_type == "UNetMidBlock2DCrossAttn":
|
531 |
+
self.mid_block = UNetMidBlock2DCrossAttn(
|
532 |
+
transformer_layers_per_block=transformer_layers_per_block[-1],
|
533 |
+
in_channels=block_out_channels[-1],
|
534 |
+
temb_channels=blocks_time_embed_dim,
|
535 |
+
dropout=dropout,
|
536 |
+
resnet_eps=norm_eps,
|
537 |
+
resnet_act_fn=act_fn,
|
538 |
+
output_scale_factor=mid_block_scale_factor,
|
539 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
540 |
+
cross_attention_dim=cross_attention_dim[-1],
|
541 |
+
num_attention_heads=num_attention_heads[-1],
|
542 |
+
resnet_groups=norm_num_groups,
|
543 |
+
dual_cross_attention=dual_cross_attention,
|
544 |
+
use_linear_projection=use_linear_projection,
|
545 |
+
upcast_attention=upcast_attention,
|
546 |
+
attention_type=attention_type,
|
547 |
+
)
|
548 |
+
elif mid_block_type == "UNetMidBlock2DSimpleCrossAttn":
|
549 |
+
raise NotImplementedError(f"Unsupport mid_block_type: {mid_block_type}")
|
550 |
+
elif mid_block_type == "UNetMidBlock2D":
|
551 |
+
self.mid_block = UNetMidBlock2D(
|
552 |
+
in_channels=block_out_channels[-1],
|
553 |
+
temb_channels=blocks_time_embed_dim,
|
554 |
+
dropout=dropout,
|
555 |
+
num_layers=0,
|
556 |
+
resnet_eps=norm_eps,
|
557 |
+
resnet_act_fn=act_fn,
|
558 |
+
output_scale_factor=mid_block_scale_factor,
|
559 |
+
resnet_groups=norm_num_groups,
|
560 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
561 |
+
add_attention=False,
|
562 |
+
)
|
563 |
+
elif mid_block_type is None:
|
564 |
+
self.mid_block = None
|
565 |
+
else:
|
566 |
+
raise ValueError(f"unknown mid_block_type : {mid_block_type}")
|
567 |
+
|
568 |
+
# count how many layers upsample the images
|
569 |
+
self.num_upsamplers = 0
|
570 |
+
|
571 |
+
# up
|
572 |
+
reversed_block_out_channels = list(reversed(block_out_channels))
|
573 |
+
reversed_num_attention_heads = list(reversed(num_attention_heads))
|
574 |
+
reversed_layers_per_block = list(reversed(layers_per_block))
|
575 |
+
reversed_cross_attention_dim = list(reversed(cross_attention_dim))
|
576 |
+
reversed_transformer_layers_per_block = (
|
577 |
+
list(reversed(transformer_layers_per_block))
|
578 |
+
if reverse_transformer_layers_per_block is None
|
579 |
+
else reverse_transformer_layers_per_block
|
580 |
+
)
|
581 |
+
only_cross_attention = list(reversed(only_cross_attention))
|
582 |
+
|
583 |
+
output_channel = reversed_block_out_channels[0]
|
584 |
+
for i, up_block_type in enumerate(up_block_types):
|
585 |
+
is_final_block = i == len(block_out_channels) - 1
|
586 |
+
|
587 |
+
prev_output_channel = output_channel
|
588 |
+
output_channel = reversed_block_out_channels[i]
|
589 |
+
input_channel = reversed_block_out_channels[
|
590 |
+
min(i + 1, len(block_out_channels) - 1)
|
591 |
+
]
|
592 |
+
|
593 |
+
# add upsample block for all BUT final layer
|
594 |
+
if not is_final_block:
|
595 |
+
add_upsample = True
|
596 |
+
self.num_upsamplers += 1
|
597 |
+
else:
|
598 |
+
add_upsample = False
|
599 |
+
|
600 |
+
up_block = get_up_block(
|
601 |
+
up_block_type,
|
602 |
+
num_layers=reversed_layers_per_block[i] + 1,
|
603 |
+
transformer_layers_per_block=reversed_transformer_layers_per_block[i],
|
604 |
+
in_channels=input_channel,
|
605 |
+
out_channels=output_channel,
|
606 |
+
prev_output_channel=prev_output_channel,
|
607 |
+
temb_channels=blocks_time_embed_dim,
|
608 |
+
add_upsample=add_upsample,
|
609 |
+
resnet_eps=norm_eps,
|
610 |
+
resnet_act_fn=act_fn,
|
611 |
+
resolution_idx=i,
|
612 |
+
resnet_groups=norm_num_groups,
|
613 |
+
cross_attention_dim=reversed_cross_attention_dim[i],
|
614 |
+
num_attention_heads=reversed_num_attention_heads[i],
|
615 |
+
dual_cross_attention=dual_cross_attention,
|
616 |
+
use_linear_projection=use_linear_projection,
|
617 |
+
only_cross_attention=only_cross_attention[i],
|
618 |
+
upcast_attention=upcast_attention,
|
619 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
620 |
+
attention_type=attention_type,
|
621 |
+
resnet_skip_time_act=resnet_skip_time_act,
|
622 |
+
resnet_out_scale_factor=resnet_out_scale_factor,
|
623 |
+
cross_attention_norm=cross_attention_norm,
|
624 |
+
attention_head_dim=attention_head_dim[i]
|
625 |
+
if attention_head_dim[i] is not None
|
626 |
+
else output_channel,
|
627 |
+
dropout=dropout,
|
628 |
+
)
|
629 |
+
self.up_blocks.append(up_block)
|
630 |
+
prev_output_channel = output_channel
|
631 |
+
|
632 |
+
# out
|
633 |
+
if norm_num_groups is not None:
|
634 |
+
self.conv_norm_out = nn.GroupNorm(
|
635 |
+
num_channels=block_out_channels[0],
|
636 |
+
num_groups=norm_num_groups,
|
637 |
+
eps=norm_eps,
|
638 |
+
)
|
639 |
+
|
640 |
+
self.conv_act = get_activation(act_fn)
|
641 |
+
|
642 |
+
else:
|
643 |
+
self.conv_norm_out = None
|
644 |
+
self.conv_act = None
|
645 |
+
self.conv_norm_out = None
|
646 |
+
|
647 |
+
conv_out_padding = (conv_out_kernel - 1) // 2
|
648 |
+
# self.conv_out = nn.Conv2d(
|
649 |
+
# block_out_channels[0],
|
650 |
+
# out_channels,
|
651 |
+
# kernel_size=conv_out_kernel,
|
652 |
+
# padding=conv_out_padding,
|
653 |
+
# )
|
654 |
+
|
655 |
+
if attention_type in ["gated", "gated-text-image"]:
|
656 |
+
positive_len = 768
|
657 |
+
if isinstance(cross_attention_dim, int):
|
658 |
+
positive_len = cross_attention_dim
|
659 |
+
elif isinstance(cross_attention_dim, tuple) or isinstance(
|
660 |
+
cross_attention_dim, list
|
661 |
+
):
|
662 |
+
positive_len = cross_attention_dim[0]
|
663 |
+
|
664 |
+
feature_type = "text-only" if attention_type == "gated" else "text-image"
|
665 |
+
self.position_net = PositionNet(
|
666 |
+
positive_len=positive_len,
|
667 |
+
out_dim=cross_attention_dim,
|
668 |
+
feature_type=feature_type,
|
669 |
+
)
|
670 |
+
|
671 |
+
@property
|
672 |
+
def attn_processors(self) -> Dict[str, AttentionProcessor]:
|
673 |
+
r"""
|
674 |
+
Returns:
|
675 |
+
`dict` of attention processors: A dictionary containing all attention processors used in the model with
|
676 |
+
indexed by its weight name.
|
677 |
+
"""
|
678 |
+
# set recursively
|
679 |
+
processors = {}
|
680 |
+
|
681 |
+
def fn_recursive_add_processors(
|
682 |
+
name: str,
|
683 |
+
module: torch.nn.Module,
|
684 |
+
processors: Dict[str, AttentionProcessor],
|
685 |
+
):
|
686 |
+
if hasattr(module, "get_processor"):
|
687 |
+
processors[f"{name}.processor"] = module.get_processor(
|
688 |
+
return_deprecated_lora=True
|
689 |
+
)
|
690 |
+
|
691 |
+
for sub_name, child in module.named_children():
|
692 |
+
fn_recursive_add_processors(f"{name}.{sub_name}", child, processors)
|
693 |
+
|
694 |
+
return processors
|
695 |
+
|
696 |
+
for name, module in self.named_children():
|
697 |
+
fn_recursive_add_processors(name, module, processors)
|
698 |
+
|
699 |
+
return processors
|
700 |
+
|
701 |
+
def set_attn_processor(
|
702 |
+
self,
|
703 |
+
processor: Union[AttentionProcessor, Dict[str, AttentionProcessor]],
|
704 |
+
_remove_lora=False,
|
705 |
+
):
|
706 |
+
r"""
|
707 |
+
Sets the attention processor to use to compute attention.
|
708 |
+
|
709 |
+
Parameters:
|
710 |
+
processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`):
|
711 |
+
The instantiated processor class or a dictionary of processor classes that will be set as the processor
|
712 |
+
for **all** `Attention` layers.
|
713 |
+
|
714 |
+
If `processor` is a dict, the key needs to define the path to the corresponding cross attention
|
715 |
+
processor. This is strongly recommended when setting trainable attention processors.
|
716 |
+
|
717 |
+
"""
|
718 |
+
count = len(self.attn_processors.keys())
|
719 |
+
|
720 |
+
if isinstance(processor, dict) and len(processor) != count:
|
721 |
+
raise ValueError(
|
722 |
+
f"A dict of processors was passed, but the number of processors {len(processor)} does not match the"
|
723 |
+
f" number of attention layers: {count}. Please make sure to pass {count} processor classes."
|
724 |
+
)
|
725 |
+
|
726 |
+
def fn_recursive_attn_processor(name: str, module: torch.nn.Module, processor):
|
727 |
+
if hasattr(module, "set_processor"):
|
728 |
+
if not isinstance(processor, dict):
|
729 |
+
module.set_processor(processor, _remove_lora=_remove_lora)
|
730 |
+
else:
|
731 |
+
module.set_processor(
|
732 |
+
processor.pop(f"{name}.processor"), _remove_lora=_remove_lora
|
733 |
+
)
|
734 |
+
|
735 |
+
for sub_name, child in module.named_children():
|
736 |
+
fn_recursive_attn_processor(f"{name}.{sub_name}", child, processor)
|
737 |
+
|
738 |
+
for name, module in self.named_children():
|
739 |
+
fn_recursive_attn_processor(name, module, processor)
|
740 |
+
|
741 |
+
def set_default_attn_processor(self):
|
742 |
+
"""
|
743 |
+
Disables custom attention processors and sets the default attention implementation.
|
744 |
+
"""
|
745 |
+
if all(
|
746 |
+
proc.__class__ in ADDED_KV_ATTENTION_PROCESSORS
|
747 |
+
for proc in self.attn_processors.values()
|
748 |
+
):
|
749 |
+
processor = AttnAddedKVProcessor()
|
750 |
+
elif all(
|
751 |
+
proc.__class__ in CROSS_ATTENTION_PROCESSORS
|
752 |
+
for proc in self.attn_processors.values()
|
753 |
+
):
|
754 |
+
processor = AttnProcessor()
|
755 |
+
else:
|
756 |
+
raise ValueError(
|
757 |
+
f"Cannot call `set_default_attn_processor` when attention processors are of type {next(iter(self.attn_processors.values()))}"
|
758 |
+
)
|
759 |
+
|
760 |
+
self.set_attn_processor(processor, _remove_lora=True)
|
761 |
+
|
762 |
+
def set_attention_slice(self, slice_size):
|
763 |
+
r"""
|
764 |
+
Enable sliced attention computation.
|
765 |
+
|
766 |
+
When this option is enabled, the attention module splits the input tensor in slices to compute attention in
|
767 |
+
several steps. This is useful for saving some memory in exchange for a small decrease in speed.
|
768 |
+
|
769 |
+
Args:
|
770 |
+
slice_size (`str` or `int` or `list(int)`, *optional*, defaults to `"auto"`):
|
771 |
+
When `"auto"`, input to the attention heads is halved, so attention is computed in two steps. If
|
772 |
+
`"max"`, maximum amount of memory is saved by running only one slice at a time. If a number is
|
773 |
+
provided, uses as many slices as `attention_head_dim // slice_size`. In this case, `attention_head_dim`
|
774 |
+
must be a multiple of `slice_size`.
|
775 |
+
"""
|
776 |
+
sliceable_head_dims = []
|
777 |
+
|
778 |
+
def fn_recursive_retrieve_sliceable_dims(module: torch.nn.Module):
|
779 |
+
if hasattr(module, "set_attention_slice"):
|
780 |
+
sliceable_head_dims.append(module.sliceable_head_dim)
|
781 |
+
|
782 |
+
for child in module.children():
|
783 |
+
fn_recursive_retrieve_sliceable_dims(child)
|
784 |
+
|
785 |
+
# retrieve number of attention layers
|
786 |
+
for module in self.children():
|
787 |
+
fn_recursive_retrieve_sliceable_dims(module)
|
788 |
+
|
789 |
+
num_sliceable_layers = len(sliceable_head_dims)
|
790 |
+
|
791 |
+
if slice_size == "auto":
|
792 |
+
# half the attention head size is usually a good trade-off between
|
793 |
+
# speed and memory
|
794 |
+
slice_size = [dim // 2 for dim in sliceable_head_dims]
|
795 |
+
elif slice_size == "max":
|
796 |
+
# make smallest slice possible
|
797 |
+
slice_size = num_sliceable_layers * [1]
|
798 |
+
|
799 |
+
slice_size = (
|
800 |
+
num_sliceable_layers * [slice_size]
|
801 |
+
if not isinstance(slice_size, list)
|
802 |
+
else slice_size
|
803 |
+
)
|
804 |
+
|
805 |
+
if len(slice_size) != len(sliceable_head_dims):
|
806 |
+
raise ValueError(
|
807 |
+
f"You have provided {len(slice_size)}, but {self.config} has {len(sliceable_head_dims)} different"
|
808 |
+
f" attention layers. Make sure to match `len(slice_size)` to be {len(sliceable_head_dims)}."
|
809 |
+
)
|
810 |
+
|
811 |
+
for i in range(len(slice_size)):
|
812 |
+
size = slice_size[i]
|
813 |
+
dim = sliceable_head_dims[i]
|
814 |
+
if size is not None and size > dim:
|
815 |
+
raise ValueError(f"size {size} has to be smaller or equal to {dim}.")
|
816 |
+
|
817 |
+
# Recursively walk through all the children.
|
818 |
+
# Any children which exposes the set_attention_slice method
|
819 |
+
# gets the message
|
820 |
+
def fn_recursive_set_attention_slice(
|
821 |
+
module: torch.nn.Module, slice_size: List[int]
|
822 |
+
):
|
823 |
+
if hasattr(module, "set_attention_slice"):
|
824 |
+
module.set_attention_slice(slice_size.pop())
|
825 |
+
|
826 |
+
for child in module.children():
|
827 |
+
fn_recursive_set_attention_slice(child, slice_size)
|
828 |
+
|
829 |
+
reversed_slice_size = list(reversed(slice_size))
|
830 |
+
for module in self.children():
|
831 |
+
fn_recursive_set_attention_slice(module, reversed_slice_size)
|
832 |
+
|
833 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
834 |
+
if hasattr(module, "gradient_checkpointing"):
|
835 |
+
module.gradient_checkpointing = value
|
836 |
+
|
837 |
+
def enable_freeu(self, s1, s2, b1, b2):
|
838 |
+
r"""Enables the FreeU mechanism from https://arxiv.org/abs/2309.11497.
|
839 |
+
|
840 |
+
The suffixes after the scaling factors represent the stage blocks where they are being applied.
|
841 |
+
|
842 |
+
Please refer to the [official repository](https://github.com/ChenyangSi/FreeU) for combinations of values that
|
843 |
+
are known to work well for different pipelines such as Stable Diffusion v1, v2, and Stable Diffusion XL.
|
844 |
+
|
845 |
+
Args:
|
846 |
+
s1 (`float`):
|
847 |
+
Scaling factor for stage 1 to attenuate the contributions of the skip features. This is done to
|
848 |
+
mitigate the "oversmoothing effect" in the enhanced denoising process.
|
849 |
+
s2 (`float`):
|
850 |
+
Scaling factor for stage 2 to attenuate the contributions of the skip features. This is done to
|
851 |
+
mitigate the "oversmoothing effect" in the enhanced denoising process.
|
852 |
+
b1 (`float`): Scaling factor for stage 1 to amplify the contributions of backbone features.
|
853 |
+
b2 (`float`): Scaling factor for stage 2 to amplify the contributions of backbone features.
|
854 |
+
"""
|
855 |
+
for i, upsample_block in enumerate(self.up_blocks):
|
856 |
+
setattr(upsample_block, "s1", s1)
|
857 |
+
setattr(upsample_block, "s2", s2)
|
858 |
+
setattr(upsample_block, "b1", b1)
|
859 |
+
setattr(upsample_block, "b2", b2)
|
860 |
+
|
861 |
+
def disable_freeu(self):
|
862 |
+
"""Disables the FreeU mechanism."""
|
863 |
+
freeu_keys = {"s1", "s2", "b1", "b2"}
|
864 |
+
for i, upsample_block in enumerate(self.up_blocks):
|
865 |
+
for k in freeu_keys:
|
866 |
+
if (
|
867 |
+
hasattr(upsample_block, k)
|
868 |
+
or getattr(upsample_block, k, None) is not None
|
869 |
+
):
|
870 |
+
setattr(upsample_block, k, None)
|
871 |
+
|
872 |
+
def forward(
|
873 |
+
self,
|
874 |
+
sample: torch.FloatTensor,
|
875 |
+
timestep: Union[torch.Tensor, float, int],
|
876 |
+
encoder_hidden_states: torch.Tensor,
|
877 |
+
class_labels: Optional[torch.Tensor] = None,
|
878 |
+
timestep_cond: Optional[torch.Tensor] = None,
|
879 |
+
attention_mask: Optional[torch.Tensor] = None,
|
880 |
+
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
|
881 |
+
added_cond_kwargs: Optional[Dict[str, torch.Tensor]] = None,
|
882 |
+
down_block_additional_residuals: Optional[Tuple[torch.Tensor]] = None,
|
883 |
+
mid_block_additional_residual: Optional[torch.Tensor] = None,
|
884 |
+
down_intrablock_additional_residuals: Optional[Tuple[torch.Tensor]] = None,
|
885 |
+
encoder_attention_mask: Optional[torch.Tensor] = None,
|
886 |
+
return_dict: bool = True,
|
887 |
+
) -> Union[UNet2DConditionOutput, Tuple]:
|
888 |
+
r"""
|
889 |
+
The [`UNet2DConditionModel`] forward method.
|
890 |
+
|
891 |
+
Args:
|
892 |
+
sample (`torch.FloatTensor`):
|
893 |
+
The noisy input tensor with the following shape `(batch, channel, height, width)`.
|
894 |
+
timestep (`torch.FloatTensor` or `float` or `int`): The number of timesteps to denoise an input.
|
895 |
+
encoder_hidden_states (`torch.FloatTensor`):
|
896 |
+
The encoder hidden states with shape `(batch, sequence_length, feature_dim)`.
|
897 |
+
class_labels (`torch.Tensor`, *optional*, defaults to `None`):
|
898 |
+
Optional class labels for conditioning. Their embeddings will be summed with the timestep embeddings.
|
899 |
+
timestep_cond: (`torch.Tensor`, *optional*, defaults to `None`):
|
900 |
+
Conditional embeddings for timestep. If provided, the embeddings will be summed with the samples passed
|
901 |
+
through the `self.time_embedding` layer to obtain the timestep embeddings.
|
902 |
+
attention_mask (`torch.Tensor`, *optional*, defaults to `None`):
|
903 |
+
An attention mask of shape `(batch, key_tokens)` is applied to `encoder_hidden_states`. If `1` the mask
|
904 |
+
is kept, otherwise if `0` it is discarded. Mask will be converted into a bias, which adds large
|
905 |
+
negative values to the attention scores corresponding to "discard" tokens.
|
906 |
+
cross_attention_kwargs (`dict`, *optional*):
|
907 |
+
A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
|
908 |
+
`self.processor` in
|
909 |
+
[diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py).
|
910 |
+
added_cond_kwargs: (`dict`, *optional*):
|
911 |
+
A kwargs dictionary containing additional embeddings that if specified are added to the embeddings that
|
912 |
+
are passed along to the UNet blocks.
|
913 |
+
down_block_additional_residuals: (`tuple` of `torch.Tensor`, *optional*):
|
914 |
+
A tuple of tensors that if specified are added to the residuals of down unet blocks.
|
915 |
+
mid_block_additional_residual: (`torch.Tensor`, *optional*):
|
916 |
+
A tensor that if specified is added to the residual of the middle unet block.
|
917 |
+
encoder_attention_mask (`torch.Tensor`):
|
918 |
+
A cross-attention mask of shape `(batch, sequence_length)` is applied to `encoder_hidden_states`. If
|
919 |
+
`True` the mask is kept, otherwise if `False` it is discarded. Mask will be converted into a bias,
|
920 |
+
which adds large negative values to the attention scores corresponding to "discard" tokens.
|
921 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
922 |
+
Whether or not to return a [`~models.unet_2d_condition.UNet2DConditionOutput`] instead of a plain
|
923 |
+
tuple.
|
924 |
+
cross_attention_kwargs (`dict`, *optional*):
|
925 |
+
A kwargs dictionary that if specified is passed along to the [`AttnProcessor`].
|
926 |
+
added_cond_kwargs: (`dict`, *optional*):
|
927 |
+
A kwargs dictionary containin additional embeddings that if specified are added to the embeddings that
|
928 |
+
are passed along to the UNet blocks.
|
929 |
+
down_block_additional_residuals (`tuple` of `torch.Tensor`, *optional*):
|
930 |
+
additional residuals to be added to UNet long skip connections from down blocks to up blocks for
|
931 |
+
example from ControlNet side model(s)
|
932 |
+
mid_block_additional_residual (`torch.Tensor`, *optional*):
|
933 |
+
additional residual to be added to UNet mid block output, for example from ControlNet side model
|
934 |
+
down_intrablock_additional_residuals (`tuple` of `torch.Tensor`, *optional*):
|
935 |
+
additional residuals to be added within UNet down blocks, for example from T2I-Adapter side model(s)
|
936 |
+
|
937 |
+
Returns:
|
938 |
+
[`~models.unet_2d_condition.UNet2DConditionOutput`] or `tuple`:
|
939 |
+
If `return_dict` is True, an [`~models.unet_2d_condition.UNet2DConditionOutput`] is returned, otherwise
|
940 |
+
a `tuple` is returned where the first element is the sample tensor.
|
941 |
+
"""
|
942 |
+
# By default samples have to be AT least a multiple of the overall upsampling factor.
|
943 |
+
# The overall upsampling factor is equal to 2 ** (# num of upsampling layers).
|
944 |
+
# However, the upsampling interpolation output size can be forced to fit any upsampling size
|
945 |
+
# on the fly if necessary.
|
946 |
+
default_overall_up_factor = 2**self.num_upsamplers
|
947 |
+
|
948 |
+
# upsample size should be forwarded when sample is not a multiple of `default_overall_up_factor`
|
949 |
+
forward_upsample_size = False
|
950 |
+
upsample_size = None
|
951 |
+
|
952 |
+
for dim in sample.shape[-2:]:
|
953 |
+
if dim % default_overall_up_factor != 0:
|
954 |
+
# Forward upsample size to force interpolation output size.
|
955 |
+
forward_upsample_size = True
|
956 |
+
break
|
957 |
+
|
958 |
+
# ensure attention_mask is a bias, and give it a singleton query_tokens dimension
|
959 |
+
# expects mask of shape:
|
960 |
+
# [batch, key_tokens]
|
961 |
+
# adds singleton query_tokens dimension:
|
962 |
+
# [batch, 1, key_tokens]
|
963 |
+
# this helps to broadcast it as a bias over attention scores, which will be in one of the following shapes:
|
964 |
+
# [batch, heads, query_tokens, key_tokens] (e.g. torch sdp attn)
|
965 |
+
# [batch * heads, query_tokens, key_tokens] (e.g. xformers or classic attn)
|
966 |
+
if attention_mask is not None:
|
967 |
+
# assume that mask is expressed as:
|
968 |
+
# (1 = keep, 0 = discard)
|
969 |
+
# convert mask into a bias that can be added to attention scores:
|
970 |
+
# (keep = +0, discard = -10000.0)
|
971 |
+
attention_mask = (1 - attention_mask.to(sample.dtype)) * -10000.0
|
972 |
+
attention_mask = attention_mask.unsqueeze(1)
|
973 |
+
|
974 |
+
# convert encoder_attention_mask to a bias the same way we do for attention_mask
|
975 |
+
if encoder_attention_mask is not None:
|
976 |
+
encoder_attention_mask = (
|
977 |
+
1 - encoder_attention_mask.to(sample.dtype)
|
978 |
+
) * -10000.0
|
979 |
+
encoder_attention_mask = encoder_attention_mask.unsqueeze(1)
|
980 |
+
|
981 |
+
# 0. center input if necessary
|
982 |
+
if self.config.center_input_sample:
|
983 |
+
sample = 2 * sample - 1.0
|
984 |
+
|
985 |
+
# 1. time
|
986 |
+
timesteps = timestep
|
987 |
+
if not torch.is_tensor(timesteps):
|
988 |
+
# TODO: this requires sync between CPU and GPU. So try to pass timesteps as tensors if you can
|
989 |
+
# This would be a good case for the `match` statement (Python 3.10+)
|
990 |
+
is_mps = sample.device.type == "mps"
|
991 |
+
if isinstance(timestep, float):
|
992 |
+
dtype = torch.float32 if is_mps else torch.float64
|
993 |
+
else:
|
994 |
+
dtype = torch.int32 if is_mps else torch.int64
|
995 |
+
timesteps = torch.tensor([timesteps], dtype=dtype, device=sample.device)
|
996 |
+
elif len(timesteps.shape) == 0:
|
997 |
+
timesteps = timesteps[None].to(sample.device)
|
998 |
+
|
999 |
+
# broadcast to batch dimension in a way that's compatible with ONNX/Core ML
|
1000 |
+
timesteps = timesteps.expand(sample.shape[0])
|
1001 |
+
|
1002 |
+
t_emb = self.time_proj(timesteps)
|
1003 |
+
|
1004 |
+
# `Timesteps` does not contain any weights and will always return f32 tensors
|
1005 |
+
# but time_embedding might actually be running in fp16. so we need to cast here.
|
1006 |
+
# there might be better ways to encapsulate this.
|
1007 |
+
t_emb = t_emb.to(dtype=sample.dtype)
|
1008 |
+
|
1009 |
+
emb = self.time_embedding(t_emb, timestep_cond)
|
1010 |
+
aug_emb = None
|
1011 |
+
|
1012 |
+
if self.class_embedding is not None:
|
1013 |
+
if class_labels is None:
|
1014 |
+
raise ValueError(
|
1015 |
+
"class_labels should be provided when num_class_embeds > 0"
|
1016 |
+
)
|
1017 |
+
|
1018 |
+
if self.config.class_embed_type == "timestep":
|
1019 |
+
class_labels = self.time_proj(class_labels)
|
1020 |
+
|
1021 |
+
# `Timesteps` does not contain any weights and will always return f32 tensors
|
1022 |
+
# there might be better ways to encapsulate this.
|
1023 |
+
class_labels = class_labels.to(dtype=sample.dtype)
|
1024 |
+
|
1025 |
+
class_emb = self.class_embedding(class_labels).to(dtype=sample.dtype)
|
1026 |
+
|
1027 |
+
if self.config.class_embeddings_concat:
|
1028 |
+
emb = torch.cat([emb, class_emb], dim=-1)
|
1029 |
+
else:
|
1030 |
+
emb = emb + class_emb
|
1031 |
+
|
1032 |
+
if self.config.addition_embed_type == "text":
|
1033 |
+
aug_emb = self.add_embedding(encoder_hidden_states)
|
1034 |
+
elif self.config.addition_embed_type == "text_image":
|
1035 |
+
# Kandinsky 2.1 - style
|
1036 |
+
if "image_embeds" not in added_cond_kwargs:
|
1037 |
+
raise ValueError(
|
1038 |
+
f"{self.__class__} has the config param `addition_embed_type` set to 'text_image' which requires the keyword argument `image_embeds` to be passed in `added_cond_kwargs`"
|
1039 |
+
)
|
1040 |
+
|
1041 |
+
image_embs = added_cond_kwargs.get("image_embeds")
|
1042 |
+
text_embs = added_cond_kwargs.get("text_embeds", encoder_hidden_states)
|
1043 |
+
aug_emb = self.add_embedding(text_embs, image_embs)
|
1044 |
+
elif self.config.addition_embed_type == "text_time":
|
1045 |
+
# SDXL - style
|
1046 |
+
if "text_embeds" not in added_cond_kwargs:
|
1047 |
+
raise ValueError(
|
1048 |
+
f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `text_embeds` to be passed in `added_cond_kwargs`"
|
1049 |
+
)
|
1050 |
+
text_embeds = added_cond_kwargs.get("text_embeds")
|
1051 |
+
if "time_ids" not in added_cond_kwargs:
|
1052 |
+
raise ValueError(
|
1053 |
+
f"{self.__class__} has the config param `addition_embed_type` set to 'text_time' which requires the keyword argument `time_ids` to be passed in `added_cond_kwargs`"
|
1054 |
+
)
|
1055 |
+
time_ids = added_cond_kwargs.get("time_ids")
|
1056 |
+
time_embeds = self.add_time_proj(time_ids.flatten())
|
1057 |
+
time_embeds = time_embeds.reshape((text_embeds.shape[0], -1))
|
1058 |
+
add_embeds = torch.concat([text_embeds, time_embeds], dim=-1)
|
1059 |
+
add_embeds = add_embeds.to(emb.dtype)
|
1060 |
+
aug_emb = self.add_embedding(add_embeds)
|
1061 |
+
elif self.config.addition_embed_type == "image":
|
1062 |
+
# Kandinsky 2.2 - style
|
1063 |
+
if "image_embeds" not in added_cond_kwargs:
|
1064 |
+
raise ValueError(
|
1065 |
+
f"{self.__class__} has the config param `addition_embed_type` set to 'image' which requires the keyword argument `image_embeds` to be passed in `added_cond_kwargs`"
|
1066 |
+
)
|
1067 |
+
image_embs = added_cond_kwargs.get("image_embeds")
|
1068 |
+
aug_emb = self.add_embedding(image_embs)
|
1069 |
+
elif self.config.addition_embed_type == "image_hint":
|
1070 |
+
# Kandinsky 2.2 - style
|
1071 |
+
if (
|
1072 |
+
"image_embeds" not in added_cond_kwargs
|
1073 |
+
or "hint" not in added_cond_kwargs
|
1074 |
+
):
|
1075 |
+
raise ValueError(
|
1076 |
+
f"{self.__class__} has the config param `addition_embed_type` set to 'image_hint' which requires the keyword arguments `image_embeds` and `hint` to be passed in `added_cond_kwargs`"
|
1077 |
+
)
|
1078 |
+
image_embs = added_cond_kwargs.get("image_embeds")
|
1079 |
+
hint = added_cond_kwargs.get("hint")
|
1080 |
+
aug_emb, hint = self.add_embedding(image_embs, hint)
|
1081 |
+
sample = torch.cat([sample, hint], dim=1)
|
1082 |
+
|
1083 |
+
emb = emb + aug_emb if aug_emb is not None else emb
|
1084 |
+
|
1085 |
+
if self.time_embed_act is not None:
|
1086 |
+
emb = self.time_embed_act(emb)
|
1087 |
+
|
1088 |
+
if (
|
1089 |
+
self.encoder_hid_proj is not None
|
1090 |
+
and self.config.encoder_hid_dim_type == "text_proj"
|
1091 |
+
):
|
1092 |
+
encoder_hidden_states = self.encoder_hid_proj(encoder_hidden_states)
|
1093 |
+
elif (
|
1094 |
+
self.encoder_hid_proj is not None
|
1095 |
+
and self.config.encoder_hid_dim_type == "text_image_proj"
|
1096 |
+
):
|
1097 |
+
# Kadinsky 2.1 - style
|
1098 |
+
if "image_embeds" not in added_cond_kwargs:
|
1099 |
+
raise ValueError(
|
1100 |
+
f"{self.__class__} has the config param `encoder_hid_dim_type` set to 'text_image_proj' which requires the keyword argument `image_embeds` to be passed in `added_conditions`"
|
1101 |
+
)
|
1102 |
+
|
1103 |
+
image_embeds = added_cond_kwargs.get("image_embeds")
|
1104 |
+
encoder_hidden_states = self.encoder_hid_proj(
|
1105 |
+
encoder_hidden_states, image_embeds
|
1106 |
+
)
|
1107 |
+
elif (
|
1108 |
+
self.encoder_hid_proj is not None
|
1109 |
+
and self.config.encoder_hid_dim_type == "image_proj"
|
1110 |
+
):
|
1111 |
+
# Kandinsky 2.2 - style
|
1112 |
+
if "image_embeds" not in added_cond_kwargs:
|
1113 |
+
raise ValueError(
|
1114 |
+
f"{self.__class__} has the config param `encoder_hid_dim_type` set to 'image_proj' which requires the keyword argument `image_embeds` to be passed in `added_conditions`"
|
1115 |
+
)
|
1116 |
+
image_embeds = added_cond_kwargs.get("image_embeds")
|
1117 |
+
encoder_hidden_states = self.encoder_hid_proj(image_embeds)
|
1118 |
+
elif (
|
1119 |
+
self.encoder_hid_proj is not None
|
1120 |
+
and self.config.encoder_hid_dim_type == "ip_image_proj"
|
1121 |
+
):
|
1122 |
+
if "image_embeds" not in added_cond_kwargs:
|
1123 |
+
raise ValueError(
|
1124 |
+
f"{self.__class__} has the config param `encoder_hid_dim_type` set to 'ip_image_proj' which requires the keyword argument `image_embeds` to be passed in `added_conditions`"
|
1125 |
+
)
|
1126 |
+
image_embeds = added_cond_kwargs.get("image_embeds")
|
1127 |
+
image_embeds = self.encoder_hid_proj(image_embeds).to(
|
1128 |
+
encoder_hidden_states.dtype
|
1129 |
+
)
|
1130 |
+
encoder_hidden_states = torch.cat(
|
1131 |
+
[encoder_hidden_states, image_embeds], dim=1
|
1132 |
+
)
|
1133 |
+
|
1134 |
+
# 2. pre-process
|
1135 |
+
sample = self.conv_in(sample)
|
1136 |
+
|
1137 |
+
# 2.5 GLIGEN position net
|
1138 |
+
if (
|
1139 |
+
cross_attention_kwargs is not None
|
1140 |
+
and cross_attention_kwargs.get("gligen", None) is not None
|
1141 |
+
):
|
1142 |
+
cross_attention_kwargs = cross_attention_kwargs.copy()
|
1143 |
+
gligen_args = cross_attention_kwargs.pop("gligen")
|
1144 |
+
cross_attention_kwargs["gligen"] = {
|
1145 |
+
"objs": self.position_net(**gligen_args)
|
1146 |
+
}
|
1147 |
+
|
1148 |
+
# 3. down
|
1149 |
+
lora_scale = (
|
1150 |
+
cross_attention_kwargs.get("scale", 1.0)
|
1151 |
+
if cross_attention_kwargs is not None
|
1152 |
+
else 1.0
|
1153 |
+
)
|
1154 |
+
if USE_PEFT_BACKEND:
|
1155 |
+
# weight the lora layers by setting `lora_scale` for each PEFT layer
|
1156 |
+
scale_lora_layers(self, lora_scale)
|
1157 |
+
|
1158 |
+
is_controlnet = (
|
1159 |
+
mid_block_additional_residual is not None
|
1160 |
+
and down_block_additional_residuals is not None
|
1161 |
+
)
|
1162 |
+
# using new arg down_intrablock_additional_residuals for T2I-Adapters, to distinguish from controlnets
|
1163 |
+
is_adapter = down_intrablock_additional_residuals is not None
|
1164 |
+
# maintain backward compatibility for legacy usage, where
|
1165 |
+
# T2I-Adapter and ControlNet both use down_block_additional_residuals arg
|
1166 |
+
# but can only use one or the other
|
1167 |
+
if (
|
1168 |
+
not is_adapter
|
1169 |
+
and mid_block_additional_residual is None
|
1170 |
+
and down_block_additional_residuals is not None
|
1171 |
+
):
|
1172 |
+
deprecate(
|
1173 |
+
"T2I should not use down_block_additional_residuals",
|
1174 |
+
"1.3.0",
|
1175 |
+
"Passing intrablock residual connections with `down_block_additional_residuals` is deprecated \
|
1176 |
+
and will be removed in diffusers 1.3.0. `down_block_additional_residuals` should only be used \
|
1177 |
+
for ControlNet. Please make sure use `down_intrablock_additional_residuals` instead. ",
|
1178 |
+
standard_warn=False,
|
1179 |
+
)
|
1180 |
+
down_intrablock_additional_residuals = down_block_additional_residuals
|
1181 |
+
is_adapter = True
|
1182 |
+
|
1183 |
+
down_block_res_samples = (sample,)
|
1184 |
+
tot_referece_features = ()
|
1185 |
+
for downsample_block in self.down_blocks:
|
1186 |
+
if (
|
1187 |
+
hasattr(downsample_block, "has_cross_attention")
|
1188 |
+
and downsample_block.has_cross_attention
|
1189 |
+
):
|
1190 |
+
# For t2i-adapter CrossAttnDownBlock2D
|
1191 |
+
additional_residuals = {}
|
1192 |
+
if is_adapter and len(down_intrablock_additional_residuals) > 0:
|
1193 |
+
additional_residuals[
|
1194 |
+
"additional_residuals"
|
1195 |
+
] = down_intrablock_additional_residuals.pop(0)
|
1196 |
+
|
1197 |
+
sample, res_samples = downsample_block(
|
1198 |
+
hidden_states=sample,
|
1199 |
+
temb=emb,
|
1200 |
+
encoder_hidden_states=encoder_hidden_states,
|
1201 |
+
attention_mask=attention_mask,
|
1202 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
1203 |
+
encoder_attention_mask=encoder_attention_mask,
|
1204 |
+
**additional_residuals,
|
1205 |
+
)
|
1206 |
+
else:
|
1207 |
+
sample, res_samples = downsample_block(
|
1208 |
+
hidden_states=sample, temb=emb, scale=lora_scale
|
1209 |
+
)
|
1210 |
+
if is_adapter and len(down_intrablock_additional_residuals) > 0:
|
1211 |
+
sample += down_intrablock_additional_residuals.pop(0)
|
1212 |
+
|
1213 |
+
down_block_res_samples += res_samples
|
1214 |
+
|
1215 |
+
if is_controlnet:
|
1216 |
+
new_down_block_res_samples = ()
|
1217 |
+
|
1218 |
+
for down_block_res_sample, down_block_additional_residual in zip(
|
1219 |
+
down_block_res_samples, down_block_additional_residuals
|
1220 |
+
):
|
1221 |
+
down_block_res_sample = (
|
1222 |
+
down_block_res_sample + down_block_additional_residual
|
1223 |
+
)
|
1224 |
+
new_down_block_res_samples = new_down_block_res_samples + (
|
1225 |
+
down_block_res_sample,
|
1226 |
+
)
|
1227 |
+
|
1228 |
+
down_block_res_samples = new_down_block_res_samples
|
1229 |
+
|
1230 |
+
# 4. mid
|
1231 |
+
if self.mid_block is not None:
|
1232 |
+
if (
|
1233 |
+
hasattr(self.mid_block, "has_cross_attention")
|
1234 |
+
and self.mid_block.has_cross_attention
|
1235 |
+
):
|
1236 |
+
sample = self.mid_block(
|
1237 |
+
sample,
|
1238 |
+
emb,
|
1239 |
+
encoder_hidden_states=encoder_hidden_states,
|
1240 |
+
attention_mask=attention_mask,
|
1241 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
1242 |
+
encoder_attention_mask=encoder_attention_mask,
|
1243 |
+
)
|
1244 |
+
else:
|
1245 |
+
sample = self.mid_block(sample, emb)
|
1246 |
+
|
1247 |
+
# To support T2I-Adapter-XL
|
1248 |
+
if (
|
1249 |
+
is_adapter
|
1250 |
+
and len(down_intrablock_additional_residuals) > 0
|
1251 |
+
and sample.shape == down_intrablock_additional_residuals[0].shape
|
1252 |
+
):
|
1253 |
+
sample += down_intrablock_additional_residuals.pop(0)
|
1254 |
+
|
1255 |
+
if is_controlnet:
|
1256 |
+
sample = sample + mid_block_additional_residual
|
1257 |
+
|
1258 |
+
# 5. up
|
1259 |
+
for i, upsample_block in enumerate(self.up_blocks):
|
1260 |
+
is_final_block = i == len(self.up_blocks) - 1
|
1261 |
+
|
1262 |
+
res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
|
1263 |
+
down_block_res_samples = down_block_res_samples[
|
1264 |
+
: -len(upsample_block.resnets)
|
1265 |
+
]
|
1266 |
+
|
1267 |
+
# if we have not reached the final block and need to forward the
|
1268 |
+
# upsample size, we do it here
|
1269 |
+
if not is_final_block and forward_upsample_size:
|
1270 |
+
upsample_size = down_block_res_samples[-1].shape[2:]
|
1271 |
+
|
1272 |
+
if (
|
1273 |
+
hasattr(upsample_block, "has_cross_attention")
|
1274 |
+
and upsample_block.has_cross_attention
|
1275 |
+
):
|
1276 |
+
sample = upsample_block(
|
1277 |
+
hidden_states=sample,
|
1278 |
+
temb=emb,
|
1279 |
+
res_hidden_states_tuple=res_samples,
|
1280 |
+
encoder_hidden_states=encoder_hidden_states,
|
1281 |
+
cross_attention_kwargs=cross_attention_kwargs,
|
1282 |
+
upsample_size=upsample_size,
|
1283 |
+
attention_mask=attention_mask,
|
1284 |
+
encoder_attention_mask=encoder_attention_mask,
|
1285 |
+
)
|
1286 |
+
else:
|
1287 |
+
sample = upsample_block(
|
1288 |
+
hidden_states=sample,
|
1289 |
+
temb=emb,
|
1290 |
+
res_hidden_states_tuple=res_samples,
|
1291 |
+
upsample_size=upsample_size,
|
1292 |
+
scale=lora_scale,
|
1293 |
+
)
|
1294 |
+
|
1295 |
+
# 6. post-process
|
1296 |
+
# if self.conv_norm_out:
|
1297 |
+
# sample = self.conv_norm_out(sample)
|
1298 |
+
# sample = self.conv_act(sample)
|
1299 |
+
# sample = self.conv_out(sample)
|
1300 |
+
|
1301 |
+
if USE_PEFT_BACKEND:
|
1302 |
+
# remove `lora_scale` from each PEFT layer
|
1303 |
+
unscale_lora_layers(self, lora_scale)
|
1304 |
+
|
1305 |
+
if not return_dict:
|
1306 |
+
return (sample,)
|
1307 |
+
|
1308 |
+
return UNet2DConditionOutput(sample=sample)
|
src/models/unet_3d.py
ADDED
@@ -0,0 +1,673 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/guoyww/AnimateDiff/blob/main/animatediff/models/unet_blocks.py
|
2 |
+
|
3 |
+
from collections import OrderedDict
|
4 |
+
from dataclasses import dataclass
|
5 |
+
import pdb
|
6 |
+
from os import PathLike
|
7 |
+
from pathlib import Path
|
8 |
+
from typing import Dict, List, Optional, Tuple, Union
|
9 |
+
|
10 |
+
import torch
|
11 |
+
import torch.nn as nn
|
12 |
+
import torch.utils.checkpoint
|
13 |
+
import torch.nn.functional as F
|
14 |
+
from diffusers.configuration_utils import ConfigMixin, register_to_config
|
15 |
+
from diffusers.models.attention_processor import AttentionProcessor
|
16 |
+
from diffusers.models.embeddings import TimestepEmbedding, Timesteps
|
17 |
+
from diffusers.models.modeling_utils import ModelMixin
|
18 |
+
from diffusers.utils import SAFETENSORS_WEIGHTS_NAME, WEIGHTS_NAME, BaseOutput, logging
|
19 |
+
from safetensors.torch import load_file
|
20 |
+
|
21 |
+
from .resnet import InflatedConv3d, InflatedGroupNorm
|
22 |
+
from .unet_3d_blocks import UNetMidBlock3DCrossAttn, get_down_block, get_up_block
|
23 |
+
|
24 |
+
logger = logging.get_logger(__name__) # pylint: disable=invalid-name
|
25 |
+
|
26 |
+
|
27 |
+
@dataclass
|
28 |
+
class UNet3DConditionOutput(BaseOutput):
|
29 |
+
sample: torch.FloatTensor
|
30 |
+
|
31 |
+
|
32 |
+
class UNet3DConditionModel(ModelMixin, ConfigMixin):
|
33 |
+
_supports_gradient_checkpointing = True
|
34 |
+
|
35 |
+
@register_to_config
|
36 |
+
def __init__(
|
37 |
+
self,
|
38 |
+
sample_size: Optional[int] = None,
|
39 |
+
in_channels: int = 4,
|
40 |
+
out_channels: int = 4,
|
41 |
+
center_input_sample: bool = False,
|
42 |
+
flip_sin_to_cos: bool = True,
|
43 |
+
freq_shift: int = 0,
|
44 |
+
down_block_types: Tuple[str] = (
|
45 |
+
"CrossAttnDownBlock3D",
|
46 |
+
"CrossAttnDownBlock3D",
|
47 |
+
"CrossAttnDownBlock3D",
|
48 |
+
"DownBlock3D",
|
49 |
+
),
|
50 |
+
mid_block_type: str = "UNetMidBlock3DCrossAttn",
|
51 |
+
up_block_types: Tuple[str] = (
|
52 |
+
"UpBlock3D",
|
53 |
+
"CrossAttnUpBlock3D",
|
54 |
+
"CrossAttnUpBlock3D",
|
55 |
+
"CrossAttnUpBlock3D",
|
56 |
+
),
|
57 |
+
only_cross_attention: Union[bool, Tuple[bool]] = False,
|
58 |
+
block_out_channels: Tuple[int] = (320, 640, 1280, 1280),
|
59 |
+
layers_per_block: int = 2,
|
60 |
+
downsample_padding: int = 1,
|
61 |
+
mid_block_scale_factor: float = 1,
|
62 |
+
act_fn: str = "silu",
|
63 |
+
norm_num_groups: int = 32,
|
64 |
+
norm_eps: float = 1e-5,
|
65 |
+
cross_attention_dim: int = 1280,
|
66 |
+
attention_head_dim: Union[int, Tuple[int]] = 8,
|
67 |
+
dual_cross_attention: bool = False,
|
68 |
+
use_linear_projection: bool = False,
|
69 |
+
class_embed_type: Optional[str] = None,
|
70 |
+
num_class_embeds: Optional[int] = None,
|
71 |
+
upcast_attention: bool = False,
|
72 |
+
resnet_time_scale_shift: str = "default",
|
73 |
+
use_inflated_groupnorm=False,
|
74 |
+
# Additional
|
75 |
+
use_motion_module=False,
|
76 |
+
motion_module_resolutions=(1, 2, 4, 8),
|
77 |
+
motion_module_mid_block=False,
|
78 |
+
motion_module_decoder_only=False,
|
79 |
+
motion_module_type=None,
|
80 |
+
motion_module_kwargs={},
|
81 |
+
unet_use_cross_frame_attention=None,
|
82 |
+
unet_use_temporal_attention=None,
|
83 |
+
):
|
84 |
+
super().__init__()
|
85 |
+
|
86 |
+
self.sample_size = sample_size
|
87 |
+
time_embed_dim = block_out_channels[0] * 4
|
88 |
+
|
89 |
+
# input
|
90 |
+
self.conv_in = InflatedConv3d(
|
91 |
+
in_channels, block_out_channels[0], kernel_size=3, padding=(1, 1)
|
92 |
+
)
|
93 |
+
|
94 |
+
# time
|
95 |
+
self.time_proj = Timesteps(block_out_channels[0], flip_sin_to_cos, freq_shift)
|
96 |
+
timestep_input_dim = block_out_channels[0]
|
97 |
+
|
98 |
+
self.time_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
|
99 |
+
|
100 |
+
# class embedding
|
101 |
+
if class_embed_type is None and num_class_embeds is not None:
|
102 |
+
self.class_embedding = nn.Embedding(num_class_embeds, time_embed_dim)
|
103 |
+
elif class_embed_type == "timestep":
|
104 |
+
self.class_embedding = TimestepEmbedding(timestep_input_dim, time_embed_dim)
|
105 |
+
elif class_embed_type == "identity":
|
106 |
+
self.class_embedding = nn.Identity(time_embed_dim, time_embed_dim)
|
107 |
+
else:
|
108 |
+
self.class_embedding = None
|
109 |
+
|
110 |
+
self.down_blocks = nn.ModuleList([])
|
111 |
+
self.mid_block = None
|
112 |
+
self.up_blocks = nn.ModuleList([])
|
113 |
+
|
114 |
+
if isinstance(only_cross_attention, bool):
|
115 |
+
only_cross_attention = [only_cross_attention] * len(down_block_types)
|
116 |
+
|
117 |
+
if isinstance(attention_head_dim, int):
|
118 |
+
attention_head_dim = (attention_head_dim,) * len(down_block_types)
|
119 |
+
|
120 |
+
# down
|
121 |
+
output_channel = block_out_channels[0]
|
122 |
+
for i, down_block_type in enumerate(down_block_types):
|
123 |
+
res = 2**i
|
124 |
+
input_channel = output_channel
|
125 |
+
output_channel = block_out_channels[i]
|
126 |
+
is_final_block = i == len(block_out_channels) - 1
|
127 |
+
|
128 |
+
down_block = get_down_block(
|
129 |
+
down_block_type,
|
130 |
+
num_layers=layers_per_block,
|
131 |
+
in_channels=input_channel,
|
132 |
+
out_channels=output_channel,
|
133 |
+
temb_channels=time_embed_dim,
|
134 |
+
add_downsample=not is_final_block,
|
135 |
+
resnet_eps=norm_eps,
|
136 |
+
resnet_act_fn=act_fn,
|
137 |
+
resnet_groups=norm_num_groups,
|
138 |
+
cross_attention_dim=cross_attention_dim,
|
139 |
+
attn_num_head_channels=attention_head_dim[i],
|
140 |
+
downsample_padding=downsample_padding,
|
141 |
+
dual_cross_attention=dual_cross_attention,
|
142 |
+
use_linear_projection=use_linear_projection,
|
143 |
+
only_cross_attention=only_cross_attention[i],
|
144 |
+
upcast_attention=upcast_attention,
|
145 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
146 |
+
unet_use_cross_frame_attention=unet_use_cross_frame_attention,
|
147 |
+
unet_use_temporal_attention=unet_use_temporal_attention,
|
148 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
149 |
+
use_motion_module=use_motion_module
|
150 |
+
and (res in motion_module_resolutions)
|
151 |
+
and (not motion_module_decoder_only),
|
152 |
+
motion_module_type=motion_module_type,
|
153 |
+
motion_module_kwargs=motion_module_kwargs,
|
154 |
+
)
|
155 |
+
self.down_blocks.append(down_block)
|
156 |
+
|
157 |
+
# mid
|
158 |
+
if mid_block_type == "UNetMidBlock3DCrossAttn":
|
159 |
+
self.mid_block = UNetMidBlock3DCrossAttn(
|
160 |
+
in_channels=block_out_channels[-1],
|
161 |
+
temb_channels=time_embed_dim,
|
162 |
+
resnet_eps=norm_eps,
|
163 |
+
resnet_act_fn=act_fn,
|
164 |
+
output_scale_factor=mid_block_scale_factor,
|
165 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
166 |
+
cross_attention_dim=cross_attention_dim,
|
167 |
+
attn_num_head_channels=attention_head_dim[-1],
|
168 |
+
resnet_groups=norm_num_groups,
|
169 |
+
dual_cross_attention=dual_cross_attention,
|
170 |
+
use_linear_projection=use_linear_projection,
|
171 |
+
upcast_attention=upcast_attention,
|
172 |
+
unet_use_cross_frame_attention=unet_use_cross_frame_attention,
|
173 |
+
unet_use_temporal_attention=unet_use_temporal_attention,
|
174 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
175 |
+
use_motion_module=use_motion_module and motion_module_mid_block,
|
176 |
+
motion_module_type=motion_module_type,
|
177 |
+
motion_module_kwargs=motion_module_kwargs,
|
178 |
+
)
|
179 |
+
else:
|
180 |
+
raise ValueError(f"unknown mid_block_type : {mid_block_type}")
|
181 |
+
|
182 |
+
# count how many layers upsample the videos
|
183 |
+
self.num_upsamplers = 0
|
184 |
+
|
185 |
+
# up
|
186 |
+
reversed_block_out_channels = list(reversed(block_out_channels))
|
187 |
+
reversed_attention_head_dim = list(reversed(attention_head_dim))
|
188 |
+
only_cross_attention = list(reversed(only_cross_attention))
|
189 |
+
output_channel = reversed_block_out_channels[0]
|
190 |
+
for i, up_block_type in enumerate(up_block_types):
|
191 |
+
res = 2 ** (3 - i)
|
192 |
+
is_final_block = i == len(block_out_channels) - 1
|
193 |
+
|
194 |
+
prev_output_channel = output_channel
|
195 |
+
output_channel = reversed_block_out_channels[i]
|
196 |
+
input_channel = reversed_block_out_channels[
|
197 |
+
min(i + 1, len(block_out_channels) - 1)
|
198 |
+
]
|
199 |
+
|
200 |
+
# add upsample block for all BUT final layer
|
201 |
+
if not is_final_block:
|
202 |
+
add_upsample = True
|
203 |
+
self.num_upsamplers += 1
|
204 |
+
else:
|
205 |
+
add_upsample = False
|
206 |
+
|
207 |
+
up_block = get_up_block(
|
208 |
+
up_block_type,
|
209 |
+
num_layers=layers_per_block + 1,
|
210 |
+
in_channels=input_channel,
|
211 |
+
out_channels=output_channel,
|
212 |
+
prev_output_channel=prev_output_channel,
|
213 |
+
temb_channels=time_embed_dim,
|
214 |
+
add_upsample=add_upsample,
|
215 |
+
resnet_eps=norm_eps,
|
216 |
+
resnet_act_fn=act_fn,
|
217 |
+
resnet_groups=norm_num_groups,
|
218 |
+
cross_attention_dim=cross_attention_dim,
|
219 |
+
attn_num_head_channels=reversed_attention_head_dim[i],
|
220 |
+
dual_cross_attention=dual_cross_attention,
|
221 |
+
use_linear_projection=use_linear_projection,
|
222 |
+
only_cross_attention=only_cross_attention[i],
|
223 |
+
upcast_attention=upcast_attention,
|
224 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
225 |
+
unet_use_cross_frame_attention=unet_use_cross_frame_attention,
|
226 |
+
unet_use_temporal_attention=unet_use_temporal_attention,
|
227 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
228 |
+
use_motion_module=use_motion_module
|
229 |
+
and (res in motion_module_resolutions),
|
230 |
+
motion_module_type=motion_module_type,
|
231 |
+
motion_module_kwargs=motion_module_kwargs,
|
232 |
+
)
|
233 |
+
self.up_blocks.append(up_block)
|
234 |
+
prev_output_channel = output_channel
|
235 |
+
|
236 |
+
# out
|
237 |
+
if use_inflated_groupnorm:
|
238 |
+
self.conv_norm_out = InflatedGroupNorm(
|
239 |
+
num_channels=block_out_channels[0],
|
240 |
+
num_groups=norm_num_groups,
|
241 |
+
eps=norm_eps,
|
242 |
+
)
|
243 |
+
else:
|
244 |
+
self.conv_norm_out = nn.GroupNorm(
|
245 |
+
num_channels=block_out_channels[0],
|
246 |
+
num_groups=norm_num_groups,
|
247 |
+
eps=norm_eps,
|
248 |
+
)
|
249 |
+
self.conv_act = nn.SiLU()
|
250 |
+
self.conv_out = InflatedConv3d(
|
251 |
+
block_out_channels[0], out_channels, kernel_size=3, padding=1
|
252 |
+
)
|
253 |
+
|
254 |
+
@property
|
255 |
+
# Copied from diffusers.models.unet_2d_condition.UNet2DConditionModel.attn_processors
|
256 |
+
def attn_processors(self) -> Dict[str, AttentionProcessor]:
|
257 |
+
r"""
|
258 |
+
Returns:
|
259 |
+
`dict` of attention processors: A dictionary containing all attention processors used in the model with
|
260 |
+
indexed by its weight name.
|
261 |
+
"""
|
262 |
+
# set recursively
|
263 |
+
processors = {}
|
264 |
+
|
265 |
+
def fn_recursive_add_processors(
|
266 |
+
name: str,
|
267 |
+
module: torch.nn.Module,
|
268 |
+
processors: Dict[str, AttentionProcessor],
|
269 |
+
):
|
270 |
+
if hasattr(module, "set_processor"):
|
271 |
+
processors[f"{name}.processor"] = module.processor
|
272 |
+
|
273 |
+
for sub_name, child in module.named_children():
|
274 |
+
if "temporal_transformer" not in sub_name:
|
275 |
+
fn_recursive_add_processors(f"{name}.{sub_name}", child, processors)
|
276 |
+
|
277 |
+
return processors
|
278 |
+
|
279 |
+
for name, module in self.named_children():
|
280 |
+
if "temporal_transformer" not in name:
|
281 |
+
fn_recursive_add_processors(name, module, processors)
|
282 |
+
|
283 |
+
return processors
|
284 |
+
|
285 |
+
def set_attention_slice(self, slice_size):
|
286 |
+
r"""
|
287 |
+
Enable sliced attention computation.
|
288 |
+
|
289 |
+
When this option is enabled, the attention module will split the input tensor in slices, to compute attention
|
290 |
+
in several steps. This is useful to save some memory in exchange for a small speed decrease.
|
291 |
+
|
292 |
+
Args:
|
293 |
+
slice_size (`str` or `int` or `list(int)`, *optional*, defaults to `"auto"`):
|
294 |
+
When `"auto"`, halves the input to the attention heads, so attention will be computed in two steps. If
|
295 |
+
`"max"`, maxium amount of memory will be saved by running only one slice at a time. If a number is
|
296 |
+
provided, uses as many slices as `attention_head_dim // slice_size`. In this case, `attention_head_dim`
|
297 |
+
must be a multiple of `slice_size`.
|
298 |
+
"""
|
299 |
+
sliceable_head_dims = []
|
300 |
+
|
301 |
+
def fn_recursive_retrieve_slicable_dims(module: torch.nn.Module):
|
302 |
+
if hasattr(module, "set_attention_slice"):
|
303 |
+
sliceable_head_dims.append(module.sliceable_head_dim)
|
304 |
+
|
305 |
+
for child in module.children():
|
306 |
+
fn_recursive_retrieve_slicable_dims(child)
|
307 |
+
|
308 |
+
# retrieve number of attention layers
|
309 |
+
for module in self.children():
|
310 |
+
fn_recursive_retrieve_slicable_dims(module)
|
311 |
+
|
312 |
+
num_slicable_layers = len(sliceable_head_dims)
|
313 |
+
|
314 |
+
if slice_size == "auto":
|
315 |
+
# half the attention head size is usually a good trade-off between
|
316 |
+
# speed and memory
|
317 |
+
slice_size = [dim // 2 for dim in sliceable_head_dims]
|
318 |
+
elif slice_size == "max":
|
319 |
+
# make smallest slice possible
|
320 |
+
slice_size = num_slicable_layers * [1]
|
321 |
+
|
322 |
+
slice_size = (
|
323 |
+
num_slicable_layers * [slice_size]
|
324 |
+
if not isinstance(slice_size, list)
|
325 |
+
else slice_size
|
326 |
+
)
|
327 |
+
|
328 |
+
if len(slice_size) != len(sliceable_head_dims):
|
329 |
+
raise ValueError(
|
330 |
+
f"You have provided {len(slice_size)}, but {self.config} has {len(sliceable_head_dims)} different"
|
331 |
+
f" attention layers. Make sure to match `len(slice_size)` to be {len(sliceable_head_dims)}."
|
332 |
+
)
|
333 |
+
|
334 |
+
for i in range(len(slice_size)):
|
335 |
+
size = slice_size[i]
|
336 |
+
dim = sliceable_head_dims[i]
|
337 |
+
if size is not None and size > dim:
|
338 |
+
raise ValueError(f"size {size} has to be smaller or equal to {dim}.")
|
339 |
+
|
340 |
+
# Recursively walk through all the children.
|
341 |
+
# Any children which exposes the set_attention_slice method
|
342 |
+
# gets the message
|
343 |
+
def fn_recursive_set_attention_slice(
|
344 |
+
module: torch.nn.Module, slice_size: List[int]
|
345 |
+
):
|
346 |
+
if hasattr(module, "set_attention_slice"):
|
347 |
+
module.set_attention_slice(slice_size.pop())
|
348 |
+
|
349 |
+
for child in module.children():
|
350 |
+
fn_recursive_set_attention_slice(child, slice_size)
|
351 |
+
|
352 |
+
reversed_slice_size = list(reversed(slice_size))
|
353 |
+
for module in self.children():
|
354 |
+
fn_recursive_set_attention_slice(module, reversed_slice_size)
|
355 |
+
|
356 |
+
def _set_gradient_checkpointing(self, module, value=False):
|
357 |
+
if hasattr(module, "gradient_checkpointing"):
|
358 |
+
module.gradient_checkpointing = value
|
359 |
+
|
360 |
+
# Copied from diffusers.models.unet_2d_condition.UNet2DConditionModel.set_attn_processor
|
361 |
+
def set_attn_processor(
|
362 |
+
self, processor: Union[AttentionProcessor, Dict[str, AttentionProcessor]]
|
363 |
+
):
|
364 |
+
r"""
|
365 |
+
Sets the attention processor to use to compute attention.
|
366 |
+
|
367 |
+
Parameters:
|
368 |
+
processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`):
|
369 |
+
The instantiated processor class or a dictionary of processor classes that will be set as the processor
|
370 |
+
for **all** `Attention` layers.
|
371 |
+
|
372 |
+
If `processor` is a dict, the key needs to define the path to the corresponding cross attention
|
373 |
+
processor. This is strongly recommended when setting trainable attention processors.
|
374 |
+
|
375 |
+
"""
|
376 |
+
count = len(self.attn_processors.keys())
|
377 |
+
|
378 |
+
if isinstance(processor, dict) and len(processor) != count:
|
379 |
+
raise ValueError(
|
380 |
+
f"A dict of processors was passed, but the number of processors {len(processor)} does not match the"
|
381 |
+
f" number of attention layers: {count}. Please make sure to pass {count} processor classes."
|
382 |
+
)
|
383 |
+
|
384 |
+
def fn_recursive_attn_processor(name: str, module: torch.nn.Module, processor):
|
385 |
+
if hasattr(module, "set_processor"):
|
386 |
+
if not isinstance(processor, dict):
|
387 |
+
module.set_processor(processor)
|
388 |
+
else:
|
389 |
+
module.set_processor(processor.pop(f"{name}.processor"))
|
390 |
+
|
391 |
+
for sub_name, child in module.named_children():
|
392 |
+
if "temporal_transformer" not in sub_name:
|
393 |
+
fn_recursive_attn_processor(f"{name}.{sub_name}", child, processor)
|
394 |
+
|
395 |
+
for name, module in self.named_children():
|
396 |
+
if "temporal_transformer" not in name:
|
397 |
+
fn_recursive_attn_processor(name, module, processor)
|
398 |
+
|
399 |
+
def forward(
|
400 |
+
self,
|
401 |
+
sample: torch.FloatTensor,
|
402 |
+
timestep: Union[torch.Tensor, float, int],
|
403 |
+
encoder_hidden_states: torch.Tensor,
|
404 |
+
class_labels: Optional[torch.Tensor] = None,
|
405 |
+
pose_cond_fea = None,
|
406 |
+
attention_mask: Optional[torch.Tensor] = None,
|
407 |
+
down_block_additional_residuals: Optional[Tuple[torch.Tensor]] = None,
|
408 |
+
mid_block_additional_residual: Optional[torch.Tensor] = None,
|
409 |
+
return_dict: bool = True,
|
410 |
+
) -> Union[UNet3DConditionOutput, Tuple]:
|
411 |
+
r"""
|
412 |
+
Args:
|
413 |
+
sample (`torch.FloatTensor`): (batch, channel, height, width) noisy inputs tensor
|
414 |
+
timestep (`torch.FloatTensor` or `float` or `int`): (batch) timesteps
|
415 |
+
encoder_hidden_states (`torch.FloatTensor`): (batch, sequence_length, feature_dim) encoder hidden states
|
416 |
+
return_dict (`bool`, *optional*, defaults to `True`):
|
417 |
+
Whether or not to return a [`models.unet_2d_condition.UNet2DConditionOutput`] instead of a plain tuple.
|
418 |
+
|
419 |
+
Returns:
|
420 |
+
[`~models.unet_2d_condition.UNet2DConditionOutput`] or `tuple`:
|
421 |
+
[`~models.unet_2d_condition.UNet2DConditionOutput`] if `return_dict` is True, otherwise a `tuple`. When
|
422 |
+
returning a tuple, the first element is the sample tensor.
|
423 |
+
"""
|
424 |
+
# By default samples have to be AT least a multiple of the overall upsampling factor.
|
425 |
+
# The overall upsampling factor is equal to 2 ** (# num of upsampling layears).
|
426 |
+
# However, the upsampling interpolation output size can be forced to fit any upsampling size
|
427 |
+
# on the fly if necessary.
|
428 |
+
default_overall_up_factor = 2**self.num_upsamplers
|
429 |
+
|
430 |
+
# upsample size should be forwarded when sample is not a multiple of `default_overall_up_factor`
|
431 |
+
forward_upsample_size = False
|
432 |
+
upsample_size = None
|
433 |
+
|
434 |
+
if any(s % default_overall_up_factor != 0 for s in sample.shape[-2:]):
|
435 |
+
logger.info("Forward upsample size to force interpolation output size.")
|
436 |
+
forward_upsample_size = True
|
437 |
+
|
438 |
+
# prepare attention_mask
|
439 |
+
if attention_mask is not None:
|
440 |
+
attention_mask = (1 - attention_mask.to(sample.dtype)) * -10000.0
|
441 |
+
attention_mask = attention_mask.unsqueeze(1)
|
442 |
+
|
443 |
+
# center input if necessary
|
444 |
+
if self.config.center_input_sample:
|
445 |
+
sample = 2 * sample - 1.0
|
446 |
+
|
447 |
+
# time
|
448 |
+
timesteps = timestep
|
449 |
+
if not torch.is_tensor(timesteps):
|
450 |
+
# This would be a good case for the `match` statement (Python 3.10+)
|
451 |
+
is_mps = sample.device.type == "mps"
|
452 |
+
if isinstance(timestep, float):
|
453 |
+
dtype = torch.float32 if is_mps else torch.float64
|
454 |
+
else:
|
455 |
+
dtype = torch.int32 if is_mps else torch.int64
|
456 |
+
timesteps = torch.tensor([timesteps], dtype=dtype, device=sample.device)
|
457 |
+
elif len(timesteps.shape) == 0:
|
458 |
+
timesteps = timesteps[None].to(sample.device)
|
459 |
+
|
460 |
+
# broadcast to batch dimension in a way that's compatible with ONNX/Core ML
|
461 |
+
timesteps = timesteps.expand(sample.shape[0])
|
462 |
+
|
463 |
+
t_emb = self.time_proj(timesteps)
|
464 |
+
|
465 |
+
# timesteps does not contain any weights and will always return f32 tensors
|
466 |
+
# but time_embedding might actually be running in fp16. so we need to cast here.
|
467 |
+
# there might be better ways to encapsulate this.
|
468 |
+
t_emb = t_emb.to(dtype=self.dtype)
|
469 |
+
emb = self.time_embedding(t_emb)
|
470 |
+
|
471 |
+
if self.class_embedding is not None:
|
472 |
+
if class_labels is None:
|
473 |
+
raise ValueError(
|
474 |
+
"class_labels should be provided when num_class_embeds > 0"
|
475 |
+
)
|
476 |
+
|
477 |
+
if self.config.class_embed_type == "timestep":
|
478 |
+
class_labels = self.time_proj(class_labels)
|
479 |
+
|
480 |
+
class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
|
481 |
+
emb = emb + class_emb
|
482 |
+
|
483 |
+
# pre-process
|
484 |
+
sample = self.conv_in(sample)
|
485 |
+
if pose_cond_fea is not None:
|
486 |
+
sample = sample + pose_cond_fea[0]
|
487 |
+
|
488 |
+
# down
|
489 |
+
down_block_res_samples = (sample,)
|
490 |
+
block_count = 1
|
491 |
+
for downsample_block in self.down_blocks:
|
492 |
+
if (
|
493 |
+
hasattr(downsample_block, "has_cross_attention")
|
494 |
+
and downsample_block.has_cross_attention
|
495 |
+
):
|
496 |
+
sample, res_samples = downsample_block(
|
497 |
+
hidden_states=sample,
|
498 |
+
temb=emb,
|
499 |
+
encoder_hidden_states=encoder_hidden_states,
|
500 |
+
attention_mask=attention_mask,
|
501 |
+
)
|
502 |
+
else:
|
503 |
+
sample, res_samples = downsample_block(
|
504 |
+
hidden_states=sample,
|
505 |
+
temb=emb,
|
506 |
+
encoder_hidden_states=encoder_hidden_states,
|
507 |
+
)
|
508 |
+
if pose_cond_fea is not None:
|
509 |
+
sample = sample + pose_cond_fea[block_count]
|
510 |
+
block_count += 1
|
511 |
+
down_block_res_samples += res_samples
|
512 |
+
|
513 |
+
if down_block_additional_residuals is not None:
|
514 |
+
new_down_block_res_samples = ()
|
515 |
+
|
516 |
+
for down_block_res_sample, down_block_additional_residual in zip(
|
517 |
+
down_block_res_samples, down_block_additional_residuals
|
518 |
+
):
|
519 |
+
down_block_res_sample = (
|
520 |
+
down_block_res_sample + down_block_additional_residual
|
521 |
+
)
|
522 |
+
new_down_block_res_samples += (down_block_res_sample,)
|
523 |
+
|
524 |
+
down_block_res_samples = new_down_block_res_samples
|
525 |
+
|
526 |
+
# mid
|
527 |
+
sample = self.mid_block(
|
528 |
+
sample,
|
529 |
+
emb,
|
530 |
+
encoder_hidden_states=encoder_hidden_states,
|
531 |
+
attention_mask=attention_mask,
|
532 |
+
)
|
533 |
+
|
534 |
+
if mid_block_additional_residual is not None:
|
535 |
+
sample = sample + mid_block_additional_residual
|
536 |
+
|
537 |
+
# up
|
538 |
+
for i, upsample_block in enumerate(self.up_blocks):
|
539 |
+
is_final_block = i == len(self.up_blocks) - 1
|
540 |
+
|
541 |
+
res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
|
542 |
+
down_block_res_samples = down_block_res_samples[
|
543 |
+
: -len(upsample_block.resnets)
|
544 |
+
]
|
545 |
+
|
546 |
+
# if we have not reached the final block and need to forward the
|
547 |
+
# upsample size, we do it here
|
548 |
+
if not is_final_block and forward_upsample_size:
|
549 |
+
upsample_size = down_block_res_samples[-1].shape[2:]
|
550 |
+
|
551 |
+
if (
|
552 |
+
hasattr(upsample_block, "has_cross_attention")
|
553 |
+
and upsample_block.has_cross_attention
|
554 |
+
):
|
555 |
+
sample = upsample_block(
|
556 |
+
hidden_states=sample,
|
557 |
+
temb=emb,
|
558 |
+
res_hidden_states_tuple=res_samples,
|
559 |
+
encoder_hidden_states=encoder_hidden_states,
|
560 |
+
upsample_size=upsample_size,
|
561 |
+
attention_mask=attention_mask,
|
562 |
+
)
|
563 |
+
else:
|
564 |
+
sample = upsample_block(
|
565 |
+
hidden_states=sample,
|
566 |
+
temb=emb,
|
567 |
+
res_hidden_states_tuple=res_samples,
|
568 |
+
upsample_size=upsample_size,
|
569 |
+
encoder_hidden_states=encoder_hidden_states,
|
570 |
+
)
|
571 |
+
|
572 |
+
# post-process
|
573 |
+
sample = self.conv_norm_out(sample)
|
574 |
+
sample = self.conv_act(sample)
|
575 |
+
sample = self.conv_out(sample)
|
576 |
+
|
577 |
+
if not return_dict:
|
578 |
+
return (sample,)
|
579 |
+
|
580 |
+
return UNet3DConditionOutput(sample=sample)
|
581 |
+
|
582 |
+
@classmethod
|
583 |
+
def from_pretrained_2d(
|
584 |
+
cls,
|
585 |
+
pretrained_model_path: PathLike,
|
586 |
+
motion_module_path: PathLike,
|
587 |
+
subfolder=None,
|
588 |
+
unet_additional_kwargs=None,
|
589 |
+
mm_zero_proj_out=False,
|
590 |
+
):
|
591 |
+
pretrained_model_path = Path(pretrained_model_path)
|
592 |
+
motion_module_path = Path(motion_module_path)
|
593 |
+
if subfolder is not None:
|
594 |
+
pretrained_model_path = pretrained_model_path.joinpath(subfolder)
|
595 |
+
logger.info(
|
596 |
+
f"loaded temporal unet's pretrained weights from {pretrained_model_path} ..."
|
597 |
+
)
|
598 |
+
|
599 |
+
config_file = pretrained_model_path / "config.json"
|
600 |
+
if not (config_file.exists() and config_file.is_file()):
|
601 |
+
raise RuntimeError(f"{config_file} does not exist or is not a file")
|
602 |
+
|
603 |
+
unet_config = cls.load_config(config_file)
|
604 |
+
unet_config["_class_name"] = cls.__name__
|
605 |
+
unet_config["down_block_types"] = [
|
606 |
+
"CrossAttnDownBlock3D",
|
607 |
+
"CrossAttnDownBlock3D",
|
608 |
+
"CrossAttnDownBlock3D",
|
609 |
+
"DownBlock3D",
|
610 |
+
]
|
611 |
+
unet_config["up_block_types"] = [
|
612 |
+
"UpBlock3D",
|
613 |
+
"CrossAttnUpBlock3D",
|
614 |
+
"CrossAttnUpBlock3D",
|
615 |
+
"CrossAttnUpBlock3D",
|
616 |
+
]
|
617 |
+
unet_config["mid_block_type"] = "UNetMidBlock3DCrossAttn"
|
618 |
+
|
619 |
+
model = cls.from_config(unet_config, **unet_additional_kwargs)
|
620 |
+
# load the vanilla weights
|
621 |
+
if pretrained_model_path.joinpath(SAFETENSORS_WEIGHTS_NAME).exists():
|
622 |
+
logger.debug(
|
623 |
+
f"loading safeTensors weights from {pretrained_model_path} ..."
|
624 |
+
)
|
625 |
+
state_dict = load_file(
|
626 |
+
pretrained_model_path.joinpath(SAFETENSORS_WEIGHTS_NAME), device="cpu"
|
627 |
+
)
|
628 |
+
|
629 |
+
elif pretrained_model_path.joinpath(WEIGHTS_NAME).exists():
|
630 |
+
logger.debug(f"loading weights from {pretrained_model_path} ...")
|
631 |
+
state_dict = torch.load(
|
632 |
+
pretrained_model_path.joinpath(WEIGHTS_NAME),
|
633 |
+
map_location="cpu",
|
634 |
+
weights_only=True,
|
635 |
+
)
|
636 |
+
else:
|
637 |
+
raise FileNotFoundError(f"no weights file found in {pretrained_model_path}")
|
638 |
+
|
639 |
+
# load the motion module weights
|
640 |
+
if motion_module_path.exists() and motion_module_path.is_file():
|
641 |
+
if motion_module_path.suffix.lower() in [".pth", ".pt", ".ckpt"]:
|
642 |
+
logger.info(f"Load motion module params from {motion_module_path}")
|
643 |
+
motion_state_dict = torch.load(
|
644 |
+
motion_module_path, map_location="cpu", weights_only=True
|
645 |
+
)
|
646 |
+
elif motion_module_path.suffix.lower() == ".safetensors":
|
647 |
+
motion_state_dict = load_file(motion_module_path, device="cpu")
|
648 |
+
else:
|
649 |
+
raise RuntimeError(
|
650 |
+
f"unknown file format for motion module weights: {motion_module_path.suffix}"
|
651 |
+
)
|
652 |
+
if mm_zero_proj_out:
|
653 |
+
logger.info(f"Zero initialize proj_out layers in motion module...")
|
654 |
+
new_motion_state_dict = OrderedDict()
|
655 |
+
for k in motion_state_dict:
|
656 |
+
if "proj_out" in k:
|
657 |
+
continue
|
658 |
+
new_motion_state_dict[k] = motion_state_dict[k]
|
659 |
+
motion_state_dict = new_motion_state_dict
|
660 |
+
|
661 |
+
# merge the state dicts
|
662 |
+
state_dict.update(motion_state_dict)
|
663 |
+
|
664 |
+
# load the weights into the model
|
665 |
+
m, u = model.load_state_dict(state_dict, strict=False)
|
666 |
+
logger.debug(f"### missing keys: {len(m)}; \n### unexpected keys: {len(u)};")
|
667 |
+
|
668 |
+
params = [
|
669 |
+
p.numel() if "temporal" in n else 0 for n, p in model.named_parameters()
|
670 |
+
]
|
671 |
+
logger.info(f"Loaded {sum(params) / 1e6}M-parameter motion module")
|
672 |
+
|
673 |
+
return model
|
src/models/unet_3d_blocks.py
ADDED
@@ -0,0 +1,861 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/unet_2d_blocks.py
|
2 |
+
|
3 |
+
import pdb
|
4 |
+
from typing import Dict, Optional
|
5 |
+
import torch
|
6 |
+
from torch import nn
|
7 |
+
|
8 |
+
from .motion_module import get_motion_module
|
9 |
+
|
10 |
+
# from .motion_module import get_motion_module
|
11 |
+
from .resnet import Downsample3D, ResnetBlock3D, Upsample3D
|
12 |
+
from .transformer_3d import Transformer3DModel
|
13 |
+
|
14 |
+
|
15 |
+
def get_down_block(
|
16 |
+
down_block_type,
|
17 |
+
num_layers,
|
18 |
+
in_channels,
|
19 |
+
out_channels,
|
20 |
+
temb_channels,
|
21 |
+
add_downsample,
|
22 |
+
resnet_eps,
|
23 |
+
resnet_act_fn,
|
24 |
+
attn_num_head_channels,
|
25 |
+
resnet_groups=None,
|
26 |
+
cross_attention_dim=None,
|
27 |
+
downsample_padding=None,
|
28 |
+
dual_cross_attention=False,
|
29 |
+
use_linear_projection=False,
|
30 |
+
only_cross_attention=False,
|
31 |
+
upcast_attention=False,
|
32 |
+
resnet_time_scale_shift="default",
|
33 |
+
unet_use_cross_frame_attention=None,
|
34 |
+
unet_use_temporal_attention=None,
|
35 |
+
use_inflated_groupnorm=None,
|
36 |
+
use_motion_module=None,
|
37 |
+
motion_module_type=None,
|
38 |
+
motion_module_kwargs=None,
|
39 |
+
):
|
40 |
+
down_block_type = (
|
41 |
+
down_block_type[7:]
|
42 |
+
if down_block_type.startswith("UNetRes")
|
43 |
+
else down_block_type
|
44 |
+
)
|
45 |
+
if down_block_type == "DownBlock3D":
|
46 |
+
return DownBlock3D(
|
47 |
+
num_layers=num_layers,
|
48 |
+
in_channels=in_channels,
|
49 |
+
out_channels=out_channels,
|
50 |
+
temb_channels=temb_channels,
|
51 |
+
add_downsample=add_downsample,
|
52 |
+
resnet_eps=resnet_eps,
|
53 |
+
resnet_act_fn=resnet_act_fn,
|
54 |
+
resnet_groups=resnet_groups,
|
55 |
+
downsample_padding=downsample_padding,
|
56 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
57 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
58 |
+
use_motion_module=use_motion_module,
|
59 |
+
motion_module_type=motion_module_type,
|
60 |
+
motion_module_kwargs=motion_module_kwargs,
|
61 |
+
)
|
62 |
+
elif down_block_type == "CrossAttnDownBlock3D":
|
63 |
+
if cross_attention_dim is None:
|
64 |
+
raise ValueError(
|
65 |
+
"cross_attention_dim must be specified for CrossAttnDownBlock3D"
|
66 |
+
)
|
67 |
+
return CrossAttnDownBlock3D(
|
68 |
+
num_layers=num_layers,
|
69 |
+
in_channels=in_channels,
|
70 |
+
out_channels=out_channels,
|
71 |
+
temb_channels=temb_channels,
|
72 |
+
add_downsample=add_downsample,
|
73 |
+
resnet_eps=resnet_eps,
|
74 |
+
resnet_act_fn=resnet_act_fn,
|
75 |
+
resnet_groups=resnet_groups,
|
76 |
+
downsample_padding=downsample_padding,
|
77 |
+
cross_attention_dim=cross_attention_dim,
|
78 |
+
attn_num_head_channels=attn_num_head_channels,
|
79 |
+
dual_cross_attention=dual_cross_attention,
|
80 |
+
use_linear_projection=use_linear_projection,
|
81 |
+
only_cross_attention=only_cross_attention,
|
82 |
+
upcast_attention=upcast_attention,
|
83 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
84 |
+
unet_use_cross_frame_attention=unet_use_cross_frame_attention,
|
85 |
+
unet_use_temporal_attention=unet_use_temporal_attention,
|
86 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
87 |
+
use_motion_module=use_motion_module,
|
88 |
+
motion_module_type=motion_module_type,
|
89 |
+
motion_module_kwargs=motion_module_kwargs,
|
90 |
+
)
|
91 |
+
raise ValueError(f"{down_block_type} does not exist.")
|
92 |
+
|
93 |
+
|
94 |
+
def get_up_block(
|
95 |
+
up_block_type,
|
96 |
+
num_layers,
|
97 |
+
in_channels,
|
98 |
+
out_channels,
|
99 |
+
prev_output_channel,
|
100 |
+
temb_channels,
|
101 |
+
add_upsample,
|
102 |
+
resnet_eps,
|
103 |
+
resnet_act_fn,
|
104 |
+
attn_num_head_channels,
|
105 |
+
resnet_groups=None,
|
106 |
+
cross_attention_dim=None,
|
107 |
+
dual_cross_attention=False,
|
108 |
+
use_linear_projection=False,
|
109 |
+
only_cross_attention=False,
|
110 |
+
upcast_attention=False,
|
111 |
+
resnet_time_scale_shift="default",
|
112 |
+
unet_use_cross_frame_attention=None,
|
113 |
+
unet_use_temporal_attention=None,
|
114 |
+
use_inflated_groupnorm=None,
|
115 |
+
use_motion_module=None,
|
116 |
+
motion_module_type=None,
|
117 |
+
motion_module_kwargs=None,
|
118 |
+
):
|
119 |
+
up_block_type = (
|
120 |
+
up_block_type[7:] if up_block_type.startswith("UNetRes") else up_block_type
|
121 |
+
)
|
122 |
+
if up_block_type == "UpBlock3D":
|
123 |
+
return UpBlock3D(
|
124 |
+
num_layers=num_layers,
|
125 |
+
in_channels=in_channels,
|
126 |
+
out_channels=out_channels,
|
127 |
+
prev_output_channel=prev_output_channel,
|
128 |
+
temb_channels=temb_channels,
|
129 |
+
add_upsample=add_upsample,
|
130 |
+
resnet_eps=resnet_eps,
|
131 |
+
resnet_act_fn=resnet_act_fn,
|
132 |
+
resnet_groups=resnet_groups,
|
133 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
134 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
135 |
+
use_motion_module=use_motion_module,
|
136 |
+
motion_module_type=motion_module_type,
|
137 |
+
motion_module_kwargs=motion_module_kwargs,
|
138 |
+
)
|
139 |
+
elif up_block_type == "CrossAttnUpBlock3D":
|
140 |
+
if cross_attention_dim is None:
|
141 |
+
raise ValueError(
|
142 |
+
"cross_attention_dim must be specified for CrossAttnUpBlock3D"
|
143 |
+
)
|
144 |
+
return CrossAttnUpBlock3D(
|
145 |
+
num_layers=num_layers,
|
146 |
+
in_channels=in_channels,
|
147 |
+
out_channels=out_channels,
|
148 |
+
prev_output_channel=prev_output_channel,
|
149 |
+
temb_channels=temb_channels,
|
150 |
+
add_upsample=add_upsample,
|
151 |
+
resnet_eps=resnet_eps,
|
152 |
+
resnet_act_fn=resnet_act_fn,
|
153 |
+
resnet_groups=resnet_groups,
|
154 |
+
cross_attention_dim=cross_attention_dim,
|
155 |
+
attn_num_head_channels=attn_num_head_channels,
|
156 |
+
dual_cross_attention=dual_cross_attention,
|
157 |
+
use_linear_projection=use_linear_projection,
|
158 |
+
only_cross_attention=only_cross_attention,
|
159 |
+
upcast_attention=upcast_attention,
|
160 |
+
resnet_time_scale_shift=resnet_time_scale_shift,
|
161 |
+
unet_use_cross_frame_attention=unet_use_cross_frame_attention,
|
162 |
+
unet_use_temporal_attention=unet_use_temporal_attention,
|
163 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
164 |
+
use_motion_module=use_motion_module,
|
165 |
+
motion_module_type=motion_module_type,
|
166 |
+
motion_module_kwargs=motion_module_kwargs,
|
167 |
+
)
|
168 |
+
raise ValueError(f"{up_block_type} does not exist.")
|
169 |
+
|
170 |
+
|
171 |
+
class UNetMidBlock3DCrossAttn(nn.Module):
|
172 |
+
def __init__(
|
173 |
+
self,
|
174 |
+
in_channels: int,
|
175 |
+
temb_channels: int,
|
176 |
+
dropout: float = 0.0,
|
177 |
+
num_layers: int = 1,
|
178 |
+
resnet_eps: float = 1e-6,
|
179 |
+
resnet_time_scale_shift: str = "default",
|
180 |
+
resnet_act_fn: str = "swish",
|
181 |
+
resnet_groups: int = 32,
|
182 |
+
resnet_pre_norm: bool = True,
|
183 |
+
attn_num_head_channels=1,
|
184 |
+
output_scale_factor=1.0,
|
185 |
+
cross_attention_dim=1280,
|
186 |
+
dual_cross_attention=False,
|
187 |
+
use_linear_projection=False,
|
188 |
+
upcast_attention=False,
|
189 |
+
unet_use_cross_frame_attention=None,
|
190 |
+
unet_use_temporal_attention=None,
|
191 |
+
use_inflated_groupnorm=None,
|
192 |
+
use_motion_module=None,
|
193 |
+
motion_module_type=None,
|
194 |
+
motion_module_kwargs=None,
|
195 |
+
):
|
196 |
+
super().__init__()
|
197 |
+
|
198 |
+
self.has_cross_attention = True
|
199 |
+
self.attn_num_head_channels = attn_num_head_channels
|
200 |
+
resnet_groups = (
|
201 |
+
resnet_groups if resnet_groups is not None else min(in_channels // 4, 32)
|
202 |
+
)
|
203 |
+
|
204 |
+
# there is always at least one resnet
|
205 |
+
resnets = [
|
206 |
+
ResnetBlock3D(
|
207 |
+
in_channels=in_channels,
|
208 |
+
out_channels=in_channels,
|
209 |
+
temb_channels=temb_channels,
|
210 |
+
eps=resnet_eps,
|
211 |
+
groups=resnet_groups,
|
212 |
+
dropout=dropout,
|
213 |
+
time_embedding_norm=resnet_time_scale_shift,
|
214 |
+
non_linearity=resnet_act_fn,
|
215 |
+
output_scale_factor=output_scale_factor,
|
216 |
+
pre_norm=resnet_pre_norm,
|
217 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
218 |
+
)
|
219 |
+
]
|
220 |
+
attentions = []
|
221 |
+
motion_modules = []
|
222 |
+
|
223 |
+
for _ in range(num_layers):
|
224 |
+
if dual_cross_attention:
|
225 |
+
raise NotImplementedError
|
226 |
+
attentions.append(
|
227 |
+
Transformer3DModel(
|
228 |
+
attn_num_head_channels,
|
229 |
+
in_channels // attn_num_head_channels,
|
230 |
+
in_channels=in_channels,
|
231 |
+
num_layers=1,
|
232 |
+
cross_attention_dim=cross_attention_dim,
|
233 |
+
norm_num_groups=resnet_groups,
|
234 |
+
use_linear_projection=use_linear_projection,
|
235 |
+
upcast_attention=upcast_attention,
|
236 |
+
unet_use_cross_frame_attention=unet_use_cross_frame_attention,
|
237 |
+
unet_use_temporal_attention=unet_use_temporal_attention,
|
238 |
+
)
|
239 |
+
)
|
240 |
+
motion_modules.append(
|
241 |
+
get_motion_module(
|
242 |
+
in_channels=in_channels,
|
243 |
+
motion_module_type=motion_module_type,
|
244 |
+
motion_module_kwargs=motion_module_kwargs,
|
245 |
+
)
|
246 |
+
if use_motion_module
|
247 |
+
else None
|
248 |
+
)
|
249 |
+
resnets.append(
|
250 |
+
ResnetBlock3D(
|
251 |
+
in_channels=in_channels,
|
252 |
+
out_channels=in_channels,
|
253 |
+
temb_channels=temb_channels,
|
254 |
+
eps=resnet_eps,
|
255 |
+
groups=resnet_groups,
|
256 |
+
dropout=dropout,
|
257 |
+
time_embedding_norm=resnet_time_scale_shift,
|
258 |
+
non_linearity=resnet_act_fn,
|
259 |
+
output_scale_factor=output_scale_factor,
|
260 |
+
pre_norm=resnet_pre_norm,
|
261 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
262 |
+
)
|
263 |
+
)
|
264 |
+
|
265 |
+
self.attentions = nn.ModuleList(attentions)
|
266 |
+
self.resnets = nn.ModuleList(resnets)
|
267 |
+
self.motion_modules = nn.ModuleList(motion_modules)
|
268 |
+
|
269 |
+
def forward(
|
270 |
+
self,
|
271 |
+
hidden_states,
|
272 |
+
temb=None,
|
273 |
+
encoder_hidden_states=None,
|
274 |
+
attention_mask=None,
|
275 |
+
):
|
276 |
+
hidden_states = self.resnets[0](hidden_states, temb)
|
277 |
+
for attn, resnet, motion_module in zip(
|
278 |
+
self.attentions, self.resnets[1:], self.motion_modules
|
279 |
+
):
|
280 |
+
hidden_states = attn(
|
281 |
+
hidden_states,
|
282 |
+
encoder_hidden_states=encoder_hidden_states,
|
283 |
+
).sample
|
284 |
+
hidden_states = (
|
285 |
+
motion_module(
|
286 |
+
hidden_states, temb, encoder_hidden_states=encoder_hidden_states
|
287 |
+
)
|
288 |
+
if motion_module is not None
|
289 |
+
else hidden_states
|
290 |
+
)
|
291 |
+
hidden_states = resnet(hidden_states, temb)
|
292 |
+
|
293 |
+
return hidden_states
|
294 |
+
|
295 |
+
|
296 |
+
class CrossAttnDownBlock3D(nn.Module):
|
297 |
+
def __init__(
|
298 |
+
self,
|
299 |
+
in_channels: int,
|
300 |
+
out_channels: int,
|
301 |
+
temb_channels: int,
|
302 |
+
dropout: float = 0.0,
|
303 |
+
num_layers: int = 1,
|
304 |
+
resnet_eps: float = 1e-6,
|
305 |
+
resnet_time_scale_shift: str = "default",
|
306 |
+
resnet_act_fn: str = "swish",
|
307 |
+
resnet_groups: int = 32,
|
308 |
+
resnet_pre_norm: bool = True,
|
309 |
+
attn_num_head_channels=1,
|
310 |
+
cross_attention_dim=1280,
|
311 |
+
output_scale_factor=1.0,
|
312 |
+
downsample_padding=1,
|
313 |
+
add_downsample=True,
|
314 |
+
dual_cross_attention=False,
|
315 |
+
use_linear_projection=False,
|
316 |
+
only_cross_attention=False,
|
317 |
+
upcast_attention=False,
|
318 |
+
unet_use_cross_frame_attention=None,
|
319 |
+
unet_use_temporal_attention=None,
|
320 |
+
use_inflated_groupnorm=None,
|
321 |
+
use_motion_module=None,
|
322 |
+
motion_module_type=None,
|
323 |
+
motion_module_kwargs=None,
|
324 |
+
):
|
325 |
+
super().__init__()
|
326 |
+
resnets = []
|
327 |
+
attentions = []
|
328 |
+
motion_modules = []
|
329 |
+
|
330 |
+
self.has_cross_attention = True
|
331 |
+
self.attn_num_head_channels = attn_num_head_channels
|
332 |
+
|
333 |
+
for i in range(num_layers):
|
334 |
+
in_channels = in_channels if i == 0 else out_channels
|
335 |
+
resnets.append(
|
336 |
+
ResnetBlock3D(
|
337 |
+
in_channels=in_channels,
|
338 |
+
out_channels=out_channels,
|
339 |
+
temb_channels=temb_channels,
|
340 |
+
eps=resnet_eps,
|
341 |
+
groups=resnet_groups,
|
342 |
+
dropout=dropout,
|
343 |
+
time_embedding_norm=resnet_time_scale_shift,
|
344 |
+
non_linearity=resnet_act_fn,
|
345 |
+
output_scale_factor=output_scale_factor,
|
346 |
+
pre_norm=resnet_pre_norm,
|
347 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
348 |
+
)
|
349 |
+
)
|
350 |
+
if dual_cross_attention:
|
351 |
+
raise NotImplementedError
|
352 |
+
attentions.append(
|
353 |
+
Transformer3DModel(
|
354 |
+
attn_num_head_channels,
|
355 |
+
out_channels // attn_num_head_channels,
|
356 |
+
in_channels=out_channels,
|
357 |
+
num_layers=1,
|
358 |
+
cross_attention_dim=cross_attention_dim,
|
359 |
+
norm_num_groups=resnet_groups,
|
360 |
+
use_linear_projection=use_linear_projection,
|
361 |
+
only_cross_attention=only_cross_attention,
|
362 |
+
upcast_attention=upcast_attention,
|
363 |
+
unet_use_cross_frame_attention=unet_use_cross_frame_attention,
|
364 |
+
unet_use_temporal_attention=unet_use_temporal_attention,
|
365 |
+
)
|
366 |
+
)
|
367 |
+
motion_modules.append(
|
368 |
+
get_motion_module(
|
369 |
+
in_channels=out_channels,
|
370 |
+
motion_module_type=motion_module_type,
|
371 |
+
motion_module_kwargs=motion_module_kwargs,
|
372 |
+
)
|
373 |
+
if use_motion_module
|
374 |
+
else None
|
375 |
+
)
|
376 |
+
|
377 |
+
self.attentions = nn.ModuleList(attentions)
|
378 |
+
self.resnets = nn.ModuleList(resnets)
|
379 |
+
self.motion_modules = nn.ModuleList(motion_modules)
|
380 |
+
|
381 |
+
if add_downsample:
|
382 |
+
self.downsamplers = nn.ModuleList(
|
383 |
+
[
|
384 |
+
Downsample3D(
|
385 |
+
out_channels,
|
386 |
+
use_conv=True,
|
387 |
+
out_channels=out_channels,
|
388 |
+
padding=downsample_padding,
|
389 |
+
name="op",
|
390 |
+
)
|
391 |
+
]
|
392 |
+
)
|
393 |
+
else:
|
394 |
+
self.downsamplers = None
|
395 |
+
|
396 |
+
self.gradient_checkpointing = False
|
397 |
+
|
398 |
+
def forward(
|
399 |
+
self,
|
400 |
+
hidden_states,
|
401 |
+
temb=None,
|
402 |
+
encoder_hidden_states=None,
|
403 |
+
attention_mask=None,
|
404 |
+
):
|
405 |
+
output_states = ()
|
406 |
+
|
407 |
+
for i, (resnet, attn, motion_module) in enumerate(
|
408 |
+
zip(self.resnets, self.attentions, self.motion_modules)
|
409 |
+
):
|
410 |
+
# self.gradient_checkpointing = False
|
411 |
+
if self.training and self.gradient_checkpointing:
|
412 |
+
|
413 |
+
def create_custom_forward(module, return_dict=None):
|
414 |
+
def custom_forward(*inputs):
|
415 |
+
if return_dict is not None:
|
416 |
+
return module(*inputs, return_dict=return_dict)
|
417 |
+
else:
|
418 |
+
return module(*inputs)
|
419 |
+
|
420 |
+
return custom_forward
|
421 |
+
|
422 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
423 |
+
create_custom_forward(resnet), hidden_states, temb
|
424 |
+
)
|
425 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
426 |
+
create_custom_forward(attn, return_dict=False),
|
427 |
+
hidden_states,
|
428 |
+
encoder_hidden_states,
|
429 |
+
)[0]
|
430 |
+
|
431 |
+
# add motion module
|
432 |
+
hidden_states = (
|
433 |
+
motion_module(
|
434 |
+
hidden_states, temb, encoder_hidden_states=encoder_hidden_states
|
435 |
+
)
|
436 |
+
if motion_module is not None
|
437 |
+
else hidden_states
|
438 |
+
)
|
439 |
+
|
440 |
+
else:
|
441 |
+
hidden_states = resnet(hidden_states, temb)
|
442 |
+
hidden_states = attn(
|
443 |
+
hidden_states,
|
444 |
+
encoder_hidden_states=encoder_hidden_states,
|
445 |
+
).sample
|
446 |
+
|
447 |
+
# add motion module
|
448 |
+
hidden_states = (
|
449 |
+
motion_module(
|
450 |
+
hidden_states, temb, encoder_hidden_states=encoder_hidden_states
|
451 |
+
)
|
452 |
+
if motion_module is not None
|
453 |
+
else hidden_states
|
454 |
+
)
|
455 |
+
|
456 |
+
output_states += (hidden_states,)
|
457 |
+
|
458 |
+
if self.downsamplers is not None:
|
459 |
+
for downsampler in self.downsamplers:
|
460 |
+
hidden_states = downsampler(hidden_states)
|
461 |
+
|
462 |
+
output_states += (hidden_states,)
|
463 |
+
|
464 |
+
return hidden_states, output_states
|
465 |
+
|
466 |
+
|
467 |
+
class DownBlock3D(nn.Module):
|
468 |
+
def __init__(
|
469 |
+
self,
|
470 |
+
in_channels: int,
|
471 |
+
out_channels: int,
|
472 |
+
temb_channels: int,
|
473 |
+
dropout: float = 0.0,
|
474 |
+
num_layers: int = 1,
|
475 |
+
resnet_eps: float = 1e-6,
|
476 |
+
resnet_time_scale_shift: str = "default",
|
477 |
+
resnet_act_fn: str = "swish",
|
478 |
+
resnet_groups: int = 32,
|
479 |
+
resnet_pre_norm: bool = True,
|
480 |
+
output_scale_factor=1.0,
|
481 |
+
add_downsample=True,
|
482 |
+
downsample_padding=1,
|
483 |
+
use_inflated_groupnorm=None,
|
484 |
+
use_motion_module=None,
|
485 |
+
motion_module_type=None,
|
486 |
+
motion_module_kwargs=None,
|
487 |
+
):
|
488 |
+
super().__init__()
|
489 |
+
resnets = []
|
490 |
+
motion_modules = []
|
491 |
+
|
492 |
+
# use_motion_module = False
|
493 |
+
for i in range(num_layers):
|
494 |
+
in_channels = in_channels if i == 0 else out_channels
|
495 |
+
resnets.append(
|
496 |
+
ResnetBlock3D(
|
497 |
+
in_channels=in_channels,
|
498 |
+
out_channels=out_channels,
|
499 |
+
temb_channels=temb_channels,
|
500 |
+
eps=resnet_eps,
|
501 |
+
groups=resnet_groups,
|
502 |
+
dropout=dropout,
|
503 |
+
time_embedding_norm=resnet_time_scale_shift,
|
504 |
+
non_linearity=resnet_act_fn,
|
505 |
+
output_scale_factor=output_scale_factor,
|
506 |
+
pre_norm=resnet_pre_norm,
|
507 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
508 |
+
)
|
509 |
+
)
|
510 |
+
motion_modules.append(
|
511 |
+
get_motion_module(
|
512 |
+
in_channels=out_channels,
|
513 |
+
motion_module_type=motion_module_type,
|
514 |
+
motion_module_kwargs=motion_module_kwargs,
|
515 |
+
)
|
516 |
+
if use_motion_module
|
517 |
+
else None
|
518 |
+
)
|
519 |
+
|
520 |
+
self.resnets = nn.ModuleList(resnets)
|
521 |
+
self.motion_modules = nn.ModuleList(motion_modules)
|
522 |
+
|
523 |
+
if add_downsample:
|
524 |
+
self.downsamplers = nn.ModuleList(
|
525 |
+
[
|
526 |
+
Downsample3D(
|
527 |
+
out_channels,
|
528 |
+
use_conv=True,
|
529 |
+
out_channels=out_channels,
|
530 |
+
padding=downsample_padding,
|
531 |
+
name="op",
|
532 |
+
)
|
533 |
+
]
|
534 |
+
)
|
535 |
+
else:
|
536 |
+
self.downsamplers = None
|
537 |
+
|
538 |
+
self.gradient_checkpointing = False
|
539 |
+
|
540 |
+
def forward(self, hidden_states, temb=None, encoder_hidden_states=None):
|
541 |
+
output_states = ()
|
542 |
+
|
543 |
+
for resnet, motion_module in zip(self.resnets, self.motion_modules):
|
544 |
+
# print(f"DownBlock3D {self.gradient_checkpointing = }")
|
545 |
+
if self.training and self.gradient_checkpointing:
|
546 |
+
|
547 |
+
def create_custom_forward(module):
|
548 |
+
def custom_forward(*inputs):
|
549 |
+
return module(*inputs)
|
550 |
+
|
551 |
+
return custom_forward
|
552 |
+
|
553 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
554 |
+
create_custom_forward(resnet), hidden_states, temb
|
555 |
+
)
|
556 |
+
if motion_module is not None:
|
557 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
558 |
+
create_custom_forward(motion_module),
|
559 |
+
hidden_states.requires_grad_(),
|
560 |
+
temb,
|
561 |
+
encoder_hidden_states,
|
562 |
+
)
|
563 |
+
else:
|
564 |
+
hidden_states = resnet(hidden_states, temb)
|
565 |
+
|
566 |
+
# add motion module
|
567 |
+
hidden_states = (
|
568 |
+
motion_module(
|
569 |
+
hidden_states, temb, encoder_hidden_states=encoder_hidden_states
|
570 |
+
)
|
571 |
+
if motion_module is not None
|
572 |
+
else hidden_states
|
573 |
+
)
|
574 |
+
|
575 |
+
output_states += (hidden_states,)
|
576 |
+
|
577 |
+
if self.downsamplers is not None:
|
578 |
+
for downsampler in self.downsamplers:
|
579 |
+
hidden_states = downsampler(hidden_states)
|
580 |
+
|
581 |
+
output_states += (hidden_states,)
|
582 |
+
|
583 |
+
return hidden_states, output_states
|
584 |
+
|
585 |
+
|
586 |
+
class CrossAttnUpBlock3D(nn.Module):
|
587 |
+
def __init__(
|
588 |
+
self,
|
589 |
+
in_channels: int,
|
590 |
+
out_channels: int,
|
591 |
+
prev_output_channel: int,
|
592 |
+
temb_channels: int,
|
593 |
+
dropout: float = 0.0,
|
594 |
+
num_layers: int = 1,
|
595 |
+
resnet_eps: float = 1e-6,
|
596 |
+
resnet_time_scale_shift: str = "default",
|
597 |
+
resnet_act_fn: str = "swish",
|
598 |
+
resnet_groups: int = 32,
|
599 |
+
resnet_pre_norm: bool = True,
|
600 |
+
attn_num_head_channels=1,
|
601 |
+
cross_attention_dim=1280,
|
602 |
+
output_scale_factor=1.0,
|
603 |
+
add_upsample=True,
|
604 |
+
dual_cross_attention=False,
|
605 |
+
use_linear_projection=False,
|
606 |
+
only_cross_attention=False,
|
607 |
+
upcast_attention=False,
|
608 |
+
unet_use_cross_frame_attention=None,
|
609 |
+
unet_use_temporal_attention=None,
|
610 |
+
use_motion_module=None,
|
611 |
+
use_inflated_groupnorm=None,
|
612 |
+
motion_module_type=None,
|
613 |
+
motion_module_kwargs=None,
|
614 |
+
):
|
615 |
+
super().__init__()
|
616 |
+
resnets = []
|
617 |
+
attentions = []
|
618 |
+
motion_modules = []
|
619 |
+
|
620 |
+
self.has_cross_attention = True
|
621 |
+
self.attn_num_head_channels = attn_num_head_channels
|
622 |
+
|
623 |
+
for i in range(num_layers):
|
624 |
+
res_skip_channels = in_channels if (i == num_layers - 1) else out_channels
|
625 |
+
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
626 |
+
|
627 |
+
resnets.append(
|
628 |
+
ResnetBlock3D(
|
629 |
+
in_channels=resnet_in_channels + res_skip_channels,
|
630 |
+
out_channels=out_channels,
|
631 |
+
temb_channels=temb_channels,
|
632 |
+
eps=resnet_eps,
|
633 |
+
groups=resnet_groups,
|
634 |
+
dropout=dropout,
|
635 |
+
time_embedding_norm=resnet_time_scale_shift,
|
636 |
+
non_linearity=resnet_act_fn,
|
637 |
+
output_scale_factor=output_scale_factor,
|
638 |
+
pre_norm=resnet_pre_norm,
|
639 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
640 |
+
)
|
641 |
+
)
|
642 |
+
if dual_cross_attention:
|
643 |
+
raise NotImplementedError
|
644 |
+
attentions.append(
|
645 |
+
Transformer3DModel(
|
646 |
+
attn_num_head_channels,
|
647 |
+
out_channels // attn_num_head_channels,
|
648 |
+
in_channels=out_channels,
|
649 |
+
num_layers=1,
|
650 |
+
cross_attention_dim=cross_attention_dim,
|
651 |
+
norm_num_groups=resnet_groups,
|
652 |
+
use_linear_projection=use_linear_projection,
|
653 |
+
only_cross_attention=only_cross_attention,
|
654 |
+
upcast_attention=upcast_attention,
|
655 |
+
unet_use_cross_frame_attention=unet_use_cross_frame_attention,
|
656 |
+
unet_use_temporal_attention=unet_use_temporal_attention,
|
657 |
+
)
|
658 |
+
)
|
659 |
+
motion_modules.append(
|
660 |
+
get_motion_module(
|
661 |
+
in_channels=out_channels,
|
662 |
+
motion_module_type=motion_module_type,
|
663 |
+
motion_module_kwargs=motion_module_kwargs,
|
664 |
+
)
|
665 |
+
if use_motion_module
|
666 |
+
else None
|
667 |
+
)
|
668 |
+
|
669 |
+
self.attentions = nn.ModuleList(attentions)
|
670 |
+
self.resnets = nn.ModuleList(resnets)
|
671 |
+
self.motion_modules = nn.ModuleList(motion_modules)
|
672 |
+
|
673 |
+
if add_upsample:
|
674 |
+
self.upsamplers = nn.ModuleList(
|
675 |
+
[Upsample3D(out_channels, use_conv=True, out_channels=out_channels)]
|
676 |
+
)
|
677 |
+
else:
|
678 |
+
self.upsamplers = None
|
679 |
+
|
680 |
+
self.gradient_checkpointing = False
|
681 |
+
|
682 |
+
def forward(
|
683 |
+
self,
|
684 |
+
hidden_states,
|
685 |
+
res_hidden_states_tuple,
|
686 |
+
temb=None,
|
687 |
+
encoder_hidden_states=None,
|
688 |
+
upsample_size=None,
|
689 |
+
attention_mask=None,
|
690 |
+
):
|
691 |
+
for i, (resnet, attn, motion_module) in enumerate(
|
692 |
+
zip(self.resnets, self.attentions, self.motion_modules)
|
693 |
+
):
|
694 |
+
# pop res hidden states
|
695 |
+
res_hidden_states = res_hidden_states_tuple[-1]
|
696 |
+
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
697 |
+
hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
|
698 |
+
|
699 |
+
if self.training and self.gradient_checkpointing:
|
700 |
+
|
701 |
+
def create_custom_forward(module, return_dict=None):
|
702 |
+
def custom_forward(*inputs):
|
703 |
+
if return_dict is not None:
|
704 |
+
return module(*inputs, return_dict=return_dict)
|
705 |
+
else:
|
706 |
+
return module(*inputs)
|
707 |
+
|
708 |
+
return custom_forward
|
709 |
+
|
710 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
711 |
+
create_custom_forward(resnet), hidden_states, temb
|
712 |
+
)
|
713 |
+
hidden_states = attn(
|
714 |
+
hidden_states,
|
715 |
+
encoder_hidden_states=encoder_hidden_states,
|
716 |
+
).sample
|
717 |
+
if motion_module is not None:
|
718 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
719 |
+
create_custom_forward(motion_module),
|
720 |
+
hidden_states.requires_grad_(),
|
721 |
+
temb,
|
722 |
+
encoder_hidden_states,
|
723 |
+
)
|
724 |
+
|
725 |
+
else:
|
726 |
+
hidden_states = resnet(hidden_states, temb)
|
727 |
+
hidden_states = attn(
|
728 |
+
hidden_states,
|
729 |
+
encoder_hidden_states=encoder_hidden_states,
|
730 |
+
).sample
|
731 |
+
|
732 |
+
# add motion module
|
733 |
+
hidden_states = (
|
734 |
+
motion_module(
|
735 |
+
hidden_states, temb, encoder_hidden_states=encoder_hidden_states
|
736 |
+
)
|
737 |
+
if motion_module is not None
|
738 |
+
else hidden_states
|
739 |
+
)
|
740 |
+
|
741 |
+
if self.upsamplers is not None:
|
742 |
+
for upsampler in self.upsamplers:
|
743 |
+
hidden_states = upsampler(hidden_states, upsample_size)
|
744 |
+
|
745 |
+
return hidden_states
|
746 |
+
|
747 |
+
class UpBlock3D(nn.Module):
|
748 |
+
def __init__(
|
749 |
+
self,
|
750 |
+
in_channels: int,
|
751 |
+
prev_output_channel: int,
|
752 |
+
out_channels: int,
|
753 |
+
temb_channels: int,
|
754 |
+
dropout: float = 0.0,
|
755 |
+
num_layers: int = 1,
|
756 |
+
resnet_eps: float = 1e-6,
|
757 |
+
resnet_time_scale_shift: str = "default",
|
758 |
+
resnet_act_fn: str = "swish",
|
759 |
+
resnet_groups: int = 32,
|
760 |
+
resnet_pre_norm: bool = True,
|
761 |
+
output_scale_factor=1.0,
|
762 |
+
add_upsample=True,
|
763 |
+
use_inflated_groupnorm=None,
|
764 |
+
use_motion_module=None,
|
765 |
+
motion_module_type=None,
|
766 |
+
motion_module_kwargs=None,
|
767 |
+
):
|
768 |
+
super().__init__()
|
769 |
+
resnets = []
|
770 |
+
motion_modules = []
|
771 |
+
|
772 |
+
# use_motion_module = False
|
773 |
+
for i in range(num_layers):
|
774 |
+
res_skip_channels = in_channels if (i == num_layers - 1) else out_channels
|
775 |
+
resnet_in_channels = prev_output_channel if i == 0 else out_channels
|
776 |
+
|
777 |
+
resnets.append(
|
778 |
+
ResnetBlock3D(
|
779 |
+
in_channels=resnet_in_channels + res_skip_channels,
|
780 |
+
out_channels=out_channels,
|
781 |
+
temb_channels=temb_channels,
|
782 |
+
eps=resnet_eps,
|
783 |
+
groups=resnet_groups,
|
784 |
+
dropout=dropout,
|
785 |
+
time_embedding_norm=resnet_time_scale_shift,
|
786 |
+
non_linearity=resnet_act_fn,
|
787 |
+
output_scale_factor=output_scale_factor,
|
788 |
+
pre_norm=resnet_pre_norm,
|
789 |
+
use_inflated_groupnorm=use_inflated_groupnorm,
|
790 |
+
)
|
791 |
+
)
|
792 |
+
motion_modules.append(
|
793 |
+
get_motion_module(
|
794 |
+
in_channels=out_channels,
|
795 |
+
motion_module_type=motion_module_type,
|
796 |
+
motion_module_kwargs=motion_module_kwargs,
|
797 |
+
)
|
798 |
+
if use_motion_module
|
799 |
+
else None
|
800 |
+
)
|
801 |
+
|
802 |
+
self.resnets = nn.ModuleList(resnets)
|
803 |
+
self.motion_modules = nn.ModuleList(motion_modules)
|
804 |
+
|
805 |
+
if add_upsample:
|
806 |
+
self.upsamplers = nn.ModuleList(
|
807 |
+
[Upsample3D(out_channels, use_conv=True, out_channels=out_channels)]
|
808 |
+
)
|
809 |
+
else:
|
810 |
+
self.upsamplers = None
|
811 |
+
|
812 |
+
self.gradient_checkpointing = False
|
813 |
+
|
814 |
+
def forward(
|
815 |
+
self,
|
816 |
+
hidden_states,
|
817 |
+
res_hidden_states_tuple,
|
818 |
+
temb=None,
|
819 |
+
upsample_size=None,
|
820 |
+
encoder_hidden_states=None,
|
821 |
+
):
|
822 |
+
for resnet, motion_module in zip(self.resnets, self.motion_modules):
|
823 |
+
# pop res hidden states
|
824 |
+
res_hidden_states = res_hidden_states_tuple[-1]
|
825 |
+
res_hidden_states_tuple = res_hidden_states_tuple[:-1]
|
826 |
+
hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
|
827 |
+
|
828 |
+
# print(f"UpBlock3D {self.gradient_checkpointing = }")
|
829 |
+
if self.training and self.gradient_checkpointing:
|
830 |
+
|
831 |
+
def create_custom_forward(module):
|
832 |
+
def custom_forward(*inputs):
|
833 |
+
return module(*inputs)
|
834 |
+
|
835 |
+
return custom_forward
|
836 |
+
|
837 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
838 |
+
create_custom_forward(resnet), hidden_states, temb
|
839 |
+
)
|
840 |
+
if motion_module is not None:
|
841 |
+
hidden_states = torch.utils.checkpoint.checkpoint(
|
842 |
+
create_custom_forward(motion_module),
|
843 |
+
hidden_states.requires_grad_(),
|
844 |
+
temb,
|
845 |
+
encoder_hidden_states,
|
846 |
+
)
|
847 |
+
else:
|
848 |
+
hidden_states = resnet(hidden_states, temb)
|
849 |
+
hidden_states = (
|
850 |
+
motion_module(
|
851 |
+
hidden_states, temb, encoder_hidden_states=encoder_hidden_states
|
852 |
+
)
|
853 |
+
if motion_module is not None
|
854 |
+
else hidden_states
|
855 |
+
)
|
856 |
+
|
857 |
+
if self.upsamplers is not None:
|
858 |
+
for upsampler in self.upsamplers:
|
859 |
+
hidden_states = upsampler(hidden_states, upsample_size)
|
860 |
+
|
861 |
+
return hidden_states
|
src/pipelines/context.py
ADDED
@@ -0,0 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# TODO: Adapted from cli
|
2 |
+
from typing import Callable, List, Optional
|
3 |
+
|
4 |
+
import numpy as np
|
5 |
+
|
6 |
+
|
7 |
+
def ordered_halving(val):
|
8 |
+
bin_str = f"{val:064b}"
|
9 |
+
bin_flip = bin_str[::-1]
|
10 |
+
as_int = int(bin_flip, 2)
|
11 |
+
|
12 |
+
return as_int / (1 << 64)
|
13 |
+
|
14 |
+
|
15 |
+
def uniform(
|
16 |
+
step: int = ...,
|
17 |
+
num_steps: Optional[int] = None,
|
18 |
+
num_frames: int = ...,
|
19 |
+
context_size: Optional[int] = None,
|
20 |
+
context_stride: int = 3,
|
21 |
+
context_overlap: int = 4,
|
22 |
+
closed_loop: bool = True,
|
23 |
+
):
|
24 |
+
if num_frames <= context_size:
|
25 |
+
yield list(range(num_frames))
|
26 |
+
return
|
27 |
+
|
28 |
+
context_stride = min(
|
29 |
+
context_stride, int(np.ceil(np.log2(num_frames / context_size))) + 1
|
30 |
+
)
|
31 |
+
|
32 |
+
for context_step in 1 << np.arange(context_stride):
|
33 |
+
pad = int(round(num_frames * ordered_halving(step)))
|
34 |
+
for j in range(
|
35 |
+
int(ordered_halving(step) * context_step) + pad,
|
36 |
+
num_frames + pad + (0 if closed_loop else -context_overlap),
|
37 |
+
(context_size * context_step - context_overlap),
|
38 |
+
):
|
39 |
+
yield [
|
40 |
+
e % num_frames
|
41 |
+
for e in range(j, j + context_size * context_step, context_step)
|
42 |
+
]
|
43 |
+
|
44 |
+
|
45 |
+
def get_context_scheduler(name: str) -> Callable:
|
46 |
+
if name == "uniform":
|
47 |
+
return uniform
|
48 |
+
else:
|
49 |
+
raise ValueError(f"Unknown context_overlap policy {name}")
|
50 |
+
|
51 |
+
|
52 |
+
def get_total_steps(
|
53 |
+
scheduler,
|
54 |
+
timesteps: List[int],
|
55 |
+
num_steps: Optional[int] = None,
|
56 |
+
num_frames: int = ...,
|
57 |
+
context_size: Optional[int] = None,
|
58 |
+
context_stride: int = 3,
|
59 |
+
context_overlap: int = 4,
|
60 |
+
closed_loop: bool = True,
|
61 |
+
):
|
62 |
+
return sum(
|
63 |
+
len(
|
64 |
+
list(
|
65 |
+
scheduler(
|
66 |
+
i,
|
67 |
+
num_steps,
|
68 |
+
num_frames,
|
69 |
+
context_size,
|
70 |
+
context_stride,
|
71 |
+
context_overlap,
|
72 |
+
)
|
73 |
+
)
|
74 |
+
)
|
75 |
+
for i in range(len(timesteps))
|
76 |
+
)
|
src/pipelines/pipeline_pose2vid_long.py
ADDED
@@ -0,0 +1,584 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Adapted from https://github.com/magic-research/magic-animate/blob/main/magicanimate/pipelines/pipeline_animation.py
|
2 |
+
import inspect
|
3 |
+
import math
|
4 |
+
from dataclasses import dataclass
|
5 |
+
from typing import Callable, List, Optional, Union
|
6 |
+
|
7 |
+
import numpy as np
|
8 |
+
import torch
|
9 |
+
import torchvision.transforms as transforms
|
10 |
+
from diffusers import DiffusionPipeline
|
11 |
+
from diffusers.image_processor import VaeImageProcessor
|
12 |
+
from diffusers.schedulers import (
|
13 |
+
DDIMScheduler,
|
14 |
+
DPMSolverMultistepScheduler,
|
15 |
+
EulerAncestralDiscreteScheduler,
|
16 |
+
EulerDiscreteScheduler,
|
17 |
+
LMSDiscreteScheduler,
|
18 |
+
PNDMScheduler,
|
19 |
+
)
|
20 |
+
from diffusers.utils import BaseOutput, deprecate, is_accelerate_available, logging
|
21 |
+
from diffusers.utils.torch_utils import randn_tensor
|
22 |
+
from einops import rearrange
|
23 |
+
from tqdm import tqdm
|
24 |
+
from transformers import CLIPImageProcessor
|
25 |
+
|
26 |
+
from src.models.mutual_self_attention import ReferenceAttentionControl
|
27 |
+
from src.pipelines.context import get_context_scheduler
|
28 |
+
from src.pipelines.utils import get_tensor_interpolation_method
|
29 |
+
|
30 |
+
|
31 |
+
@dataclass
|
32 |
+
class Pose2VideoPipelineOutput(BaseOutput):
|
33 |
+
videos: Union[torch.Tensor, np.ndarray]
|
34 |
+
|
35 |
+
|
36 |
+
class Pose2VideoPipeline(DiffusionPipeline):
|
37 |
+
_optional_components = []
|
38 |
+
|
39 |
+
def __init__(
|
40 |
+
self,
|
41 |
+
vae,
|
42 |
+
image_encoder,
|
43 |
+
reference_unet,
|
44 |
+
denoising_unet,
|
45 |
+
pose_guider,
|
46 |
+
scheduler: Union[
|
47 |
+
DDIMScheduler,
|
48 |
+
PNDMScheduler,
|
49 |
+
LMSDiscreteScheduler,
|
50 |
+
EulerDiscreteScheduler,
|
51 |
+
EulerAncestralDiscreteScheduler,
|
52 |
+
DPMSolverMultistepScheduler,
|
53 |
+
],
|
54 |
+
image_proj_model=None,
|
55 |
+
tokenizer=None,
|
56 |
+
text_encoder=None,
|
57 |
+
):
|
58 |
+
super().__init__()
|
59 |
+
|
60 |
+
self.register_modules(
|
61 |
+
vae=vae,
|
62 |
+
image_encoder=image_encoder,
|
63 |
+
reference_unet=reference_unet,
|
64 |
+
denoising_unet=denoising_unet,
|
65 |
+
pose_guider=pose_guider,
|
66 |
+
scheduler=scheduler,
|
67 |
+
image_proj_model=image_proj_model,
|
68 |
+
tokenizer=tokenizer,
|
69 |
+
text_encoder=text_encoder,
|
70 |
+
)
|
71 |
+
self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
|
72 |
+
self.clip_image_processor = CLIPImageProcessor()
|
73 |
+
self.ref_image_processor = VaeImageProcessor(
|
74 |
+
vae_scale_factor=self.vae_scale_factor, do_convert_rgb=True
|
75 |
+
)
|
76 |
+
self.cond_image_processor = VaeImageProcessor(
|
77 |
+
vae_scale_factor=self.vae_scale_factor,
|
78 |
+
do_convert_rgb=True,
|
79 |
+
do_normalize=True,
|
80 |
+
)
|
81 |
+
|
82 |
+
def enable_vae_slicing(self):
|
83 |
+
self.vae.enable_slicing()
|
84 |
+
|
85 |
+
def disable_vae_slicing(self):
|
86 |
+
self.vae.disable_slicing()
|
87 |
+
|
88 |
+
def enable_sequential_cpu_offload(self, gpu_id=0):
|
89 |
+
if is_accelerate_available():
|
90 |
+
from accelerate import cpu_offload
|
91 |
+
else:
|
92 |
+
raise ImportError("Please install accelerate via `pip install accelerate`")
|
93 |
+
|
94 |
+
device = torch.device(f"cuda:{gpu_id}")
|
95 |
+
|
96 |
+
for cpu_offloaded_model in [self.unet, self.text_encoder, self.vae]:
|
97 |
+
if cpu_offloaded_model is not None:
|
98 |
+
cpu_offload(cpu_offloaded_model, device)
|
99 |
+
|
100 |
+
@property
|
101 |
+
def _execution_device(self):
|
102 |
+
if self.device != torch.device("meta") or not hasattr(self.unet, "_hf_hook"):
|
103 |
+
return self.device
|
104 |
+
for module in self.unet.modules():
|
105 |
+
if (
|
106 |
+
hasattr(module, "_hf_hook")
|
107 |
+
and hasattr(module._hf_hook, "execution_device")
|
108 |
+
and module._hf_hook.execution_device is not None
|
109 |
+
):
|
110 |
+
return torch.device(module._hf_hook.execution_device)
|
111 |
+
return self.device
|
112 |
+
|
113 |
+
def decode_latents(self, latents):
|
114 |
+
video_length = latents.shape[2]
|
115 |
+
latents = 1 / 0.18215 * latents
|
116 |
+
latents = rearrange(latents, "b c f h w -> (b f) c h w")
|
117 |
+
# video = self.vae.decode(latents).sample
|
118 |
+
video = []
|
119 |
+
for frame_idx in tqdm(range(latents.shape[0])):
|
120 |
+
video.append(self.vae.decode(latents[frame_idx : frame_idx + 1]).sample)
|
121 |
+
video = torch.cat(video)
|
122 |
+
video = rearrange(video, "(b f) c h w -> b c f h w", f=video_length)
|
123 |
+
video = (video / 2 + 0.5).clamp(0, 1)
|
124 |
+
# we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
|
125 |
+
video = video.cpu().float().numpy()
|
126 |
+
return video
|
127 |
+
|
128 |
+
def prepare_extra_step_kwargs(self, generator, eta):
|
129 |
+
# prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
|
130 |
+
# eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
|
131 |
+
# eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
|
132 |
+
# and should be between [0, 1]
|
133 |
+
|
134 |
+
accepts_eta = "eta" in set(
|
135 |
+
inspect.signature(self.scheduler.step).parameters.keys()
|
136 |
+
)
|
137 |
+
extra_step_kwargs = {}
|
138 |
+
if accepts_eta:
|
139 |
+
extra_step_kwargs["eta"] = eta
|
140 |
+
|
141 |
+
# check if the scheduler accepts generator
|
142 |
+
accepts_generator = "generator" in set(
|
143 |
+
inspect.signature(self.scheduler.step).parameters.keys()
|
144 |
+
)
|
145 |
+
if accepts_generator:
|
146 |
+
extra_step_kwargs["generator"] = generator
|
147 |
+
return extra_step_kwargs
|
148 |
+
|
149 |
+
def prepare_latents(
|
150 |
+
self,
|
151 |
+
batch_size,
|
152 |
+
num_channels_latents,
|
153 |
+
width,
|
154 |
+
height,
|
155 |
+
video_length,
|
156 |
+
dtype,
|
157 |
+
device,
|
158 |
+
generator,
|
159 |
+
latents=None,
|
160 |
+
):
|
161 |
+
shape = (
|
162 |
+
batch_size,
|
163 |
+
num_channels_latents,
|
164 |
+
video_length,
|
165 |
+
height // self.vae_scale_factor,
|
166 |
+
width // self.vae_scale_factor,
|
167 |
+
)
|
168 |
+
if isinstance(generator, list) and len(generator) != batch_size:
|
169 |
+
raise ValueError(
|
170 |
+
f"You have passed a list of generators of length {len(generator)}, but requested an effective batch"
|
171 |
+
f" size of {batch_size}. Make sure the batch size matches the length of the generators."
|
172 |
+
)
|
173 |
+
|
174 |
+
if latents is None:
|
175 |
+
latents = randn_tensor(
|
176 |
+
shape, generator=generator, device=device, dtype=dtype
|
177 |
+
)
|
178 |
+
else:
|
179 |
+
latents = latents.to(device)
|
180 |
+
|
181 |
+
# scale the initial noise by the standard deviation required by the scheduler
|
182 |
+
latents = latents * self.scheduler.init_noise_sigma
|
183 |
+
return latents
|
184 |
+
|
185 |
+
def _encode_prompt(
|
186 |
+
self,
|
187 |
+
prompt,
|
188 |
+
device,
|
189 |
+
num_videos_per_prompt,
|
190 |
+
do_classifier_free_guidance,
|
191 |
+
negative_prompt,
|
192 |
+
):
|
193 |
+
batch_size = len(prompt) if isinstance(prompt, list) else 1
|
194 |
+
|
195 |
+
text_inputs = self.tokenizer(
|
196 |
+
prompt,
|
197 |
+
padding="max_length",
|
198 |
+
max_length=self.tokenizer.model_max_length,
|
199 |
+
truncation=True,
|
200 |
+
return_tensors="pt",
|
201 |
+
)
|
202 |
+
text_input_ids = text_inputs.input_ids
|
203 |
+
untruncated_ids = self.tokenizer(
|
204 |
+
prompt, padding="longest", return_tensors="pt"
|
205 |
+
).input_ids
|
206 |
+
|
207 |
+
if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(
|
208 |
+
text_input_ids, untruncated_ids
|
209 |
+
):
|
210 |
+
removed_text = self.tokenizer.batch_decode(
|
211 |
+
untruncated_ids[:, self.tokenizer.model_max_length - 1 : -1]
|
212 |
+
)
|
213 |
+
|
214 |
+
if (
|
215 |
+
hasattr(self.text_encoder.config, "use_attention_mask")
|
216 |
+
and self.text_encoder.config.use_attention_mask
|
217 |
+
):
|
218 |
+
attention_mask = text_inputs.attention_mask.to(device)
|
219 |
+
else:
|
220 |
+
attention_mask = None
|
221 |
+
|
222 |
+
text_embeddings = self.text_encoder(
|
223 |
+
text_input_ids.to(device),
|
224 |
+
attention_mask=attention_mask,
|
225 |
+
)
|
226 |
+
text_embeddings = text_embeddings[0]
|
227 |
+
|
228 |
+
# duplicate text embeddings for each generation per prompt, using mps friendly method
|
229 |
+
bs_embed, seq_len, _ = text_embeddings.shape
|
230 |
+
text_embeddings = text_embeddings.repeat(1, num_videos_per_prompt, 1)
|
231 |
+
text_embeddings = text_embeddings.view(
|
232 |
+
bs_embed * num_videos_per_prompt, seq_len, -1
|
233 |
+
)
|
234 |
+
|
235 |
+
# get unconditional embeddings for classifier free guidance
|
236 |
+
if do_classifier_free_guidance:
|
237 |
+
uncond_tokens: List[str]
|
238 |
+
if negative_prompt is None:
|
239 |
+
uncond_tokens = [""] * batch_size
|
240 |
+
elif type(prompt) is not type(negative_prompt):
|
241 |
+
raise TypeError(
|
242 |
+
f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
|
243 |
+
f" {type(prompt)}."
|
244 |
+
)
|
245 |
+
elif isinstance(negative_prompt, str):
|
246 |
+
uncond_tokens = [negative_prompt]
|
247 |
+
elif batch_size != len(negative_prompt):
|
248 |
+
raise ValueError(
|
249 |
+
f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
|
250 |
+
f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
|
251 |
+
" the batch size of `prompt`."
|
252 |
+
)
|
253 |
+
else:
|
254 |
+
uncond_tokens = negative_prompt
|
255 |
+
|
256 |
+
max_length = text_input_ids.shape[-1]
|
257 |
+
uncond_input = self.tokenizer(
|
258 |
+
uncond_tokens,
|
259 |
+
padding="max_length",
|
260 |
+
max_length=max_length,
|
261 |
+
truncation=True,
|
262 |
+
return_tensors="pt",
|
263 |
+
)
|
264 |
+
|
265 |
+
if (
|
266 |
+
hasattr(self.text_encoder.config, "use_attention_mask")
|
267 |
+
and self.text_encoder.config.use_attention_mask
|
268 |
+
):
|
269 |
+
attention_mask = uncond_input.attention_mask.to(device)
|
270 |
+
else:
|
271 |
+
attention_mask = None
|
272 |
+
|
273 |
+
uncond_embeddings = self.text_encoder(
|
274 |
+
uncond_input.input_ids.to(device),
|
275 |
+
attention_mask=attention_mask,
|
276 |
+
)
|
277 |
+
uncond_embeddings = uncond_embeddings[0]
|
278 |
+
|
279 |
+
# duplicate unconditional embeddings for each generation per prompt, using mps friendly method
|
280 |
+
seq_len = uncond_embeddings.shape[1]
|
281 |
+
uncond_embeddings = uncond_embeddings.repeat(1, num_videos_per_prompt, 1)
|
282 |
+
uncond_embeddings = uncond_embeddings.view(
|
283 |
+
batch_size * num_videos_per_prompt, seq_len, -1
|
284 |
+
)
|
285 |
+
|
286 |
+
# For classifier free guidance, we need to do two forward passes.
|
287 |
+
# Here we concatenate the unconditional and text embeddings into a single batch
|
288 |
+
# to avoid doing two forward passes
|
289 |
+
text_embeddings = torch.cat([uncond_embeddings, text_embeddings])
|
290 |
+
|
291 |
+
return text_embeddings
|
292 |
+
|
293 |
+
def interpolate_latents(
|
294 |
+
self, latents: torch.Tensor, interpolation_factor: int, device
|
295 |
+
):
|
296 |
+
if interpolation_factor < 2:
|
297 |
+
return latents
|
298 |
+
|
299 |
+
new_latents = torch.zeros(
|
300 |
+
(
|
301 |
+
latents.shape[0],
|
302 |
+
latents.shape[1],
|
303 |
+
((latents.shape[2] - 1) * interpolation_factor) + 1,
|
304 |
+
latents.shape[3],
|
305 |
+
latents.shape[4],
|
306 |
+
),
|
307 |
+
device=latents.device,
|
308 |
+
dtype=latents.dtype,
|
309 |
+
)
|
310 |
+
|
311 |
+
org_video_length = latents.shape[2]
|
312 |
+
rate = [i / interpolation_factor for i in range(interpolation_factor)][1:]
|
313 |
+
|
314 |
+
new_index = 0
|
315 |
+
|
316 |
+
v0 = None
|
317 |
+
v1 = None
|
318 |
+
|
319 |
+
for i0, i1 in zip(range(org_video_length), range(org_video_length)[1:]):
|
320 |
+
v0 = latents[:, :, i0, :, :]
|
321 |
+
v1 = latents[:, :, i1, :, :]
|
322 |
+
|
323 |
+
new_latents[:, :, new_index, :, :] = v0
|
324 |
+
new_index += 1
|
325 |
+
|
326 |
+
for f in rate:
|
327 |
+
v = get_tensor_interpolation_method()(
|
328 |
+
v0.to(device=device), v1.to(device=device), f
|
329 |
+
)
|
330 |
+
new_latents[:, :, new_index, :, :] = v.to(latents.device)
|
331 |
+
new_index += 1
|
332 |
+
|
333 |
+
new_latents[:, :, new_index, :, :] = v1
|
334 |
+
new_index += 1
|
335 |
+
|
336 |
+
return new_latents
|
337 |
+
|
338 |
+
@torch.no_grad()
|
339 |
+
def __call__(
|
340 |
+
self,
|
341 |
+
ref_image,
|
342 |
+
pose_images,
|
343 |
+
ref_pose_image,
|
344 |
+
width,
|
345 |
+
height,
|
346 |
+
video_length,
|
347 |
+
num_inference_steps,
|
348 |
+
guidance_scale,
|
349 |
+
num_images_per_prompt=1,
|
350 |
+
eta: float = 0.0,
|
351 |
+
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
|
352 |
+
output_type: Optional[str] = "tensor",
|
353 |
+
return_dict: bool = True,
|
354 |
+
callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
|
355 |
+
callback_steps: Optional[int] = 1,
|
356 |
+
context_schedule="uniform",
|
357 |
+
context_frames=16,
|
358 |
+
context_stride=1,
|
359 |
+
context_overlap=4,
|
360 |
+
context_batch_size=1,
|
361 |
+
interpolation_factor=1,
|
362 |
+
**kwargs,
|
363 |
+
):
|
364 |
+
# Default height and width to unet
|
365 |
+
height = height or self.unet.config.sample_size * self.vae_scale_factor
|
366 |
+
width = width or self.unet.config.sample_size * self.vae_scale_factor
|
367 |
+
|
368 |
+
device = self._execution_device
|
369 |
+
|
370 |
+
do_classifier_free_guidance = guidance_scale > 1.0
|
371 |
+
|
372 |
+
# Prepare timesteps
|
373 |
+
self.scheduler.set_timesteps(num_inference_steps, device=device)
|
374 |
+
timesteps = self.scheduler.timesteps
|
375 |
+
|
376 |
+
batch_size = 1
|
377 |
+
|
378 |
+
# Prepare clip image embeds
|
379 |
+
clip_image = self.clip_image_processor.preprocess(
|
380 |
+
ref_image.resize((224, 224)), return_tensors="pt"
|
381 |
+
).pixel_values
|
382 |
+
clip_image_embeds = self.image_encoder(
|
383 |
+
clip_image.to(device, dtype=self.image_encoder.dtype)
|
384 |
+
).image_embeds
|
385 |
+
encoder_hidden_states = clip_image_embeds.unsqueeze(1)
|
386 |
+
uncond_encoder_hidden_states = torch.zeros_like(encoder_hidden_states)
|
387 |
+
|
388 |
+
if do_classifier_free_guidance:
|
389 |
+
encoder_hidden_states = torch.cat(
|
390 |
+
[uncond_encoder_hidden_states, encoder_hidden_states], dim=0
|
391 |
+
)
|
392 |
+
|
393 |
+
reference_control_writer = ReferenceAttentionControl(
|
394 |
+
self.reference_unet,
|
395 |
+
do_classifier_free_guidance=do_classifier_free_guidance,
|
396 |
+
mode="write",
|
397 |
+
batch_size=batch_size,
|
398 |
+
fusion_blocks="full",
|
399 |
+
)
|
400 |
+
reference_control_reader = ReferenceAttentionControl(
|
401 |
+
self.denoising_unet,
|
402 |
+
do_classifier_free_guidance=do_classifier_free_guidance,
|
403 |
+
mode="read",
|
404 |
+
batch_size=batch_size,
|
405 |
+
fusion_blocks="full",
|
406 |
+
)
|
407 |
+
|
408 |
+
num_channels_latents = self.denoising_unet.in_channels
|
409 |
+
latents = self.prepare_latents(
|
410 |
+
batch_size * num_images_per_prompt,
|
411 |
+
num_channels_latents,
|
412 |
+
width,
|
413 |
+
height,
|
414 |
+
video_length,
|
415 |
+
clip_image_embeds.dtype,
|
416 |
+
device,
|
417 |
+
generator,
|
418 |
+
)
|
419 |
+
|
420 |
+
# Prepare extra step kwargs.
|
421 |
+
extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
|
422 |
+
|
423 |
+
# Prepare ref image latents
|
424 |
+
ref_image_tensor = self.ref_image_processor.preprocess(
|
425 |
+
ref_image, height=height, width=width
|
426 |
+
) # (bs, c, width, height)
|
427 |
+
ref_image_tensor = ref_image_tensor.to(
|
428 |
+
dtype=self.vae.dtype, device=self.vae.device
|
429 |
+
)
|
430 |
+
ref_image_latents = self.vae.encode(ref_image_tensor).latent_dist.mean
|
431 |
+
ref_image_latents = ref_image_latents * 0.18215 # (b, 4, h, w)
|
432 |
+
|
433 |
+
# Prepare a list of pose condition images
|
434 |
+
pose_cond_tensor_list = []
|
435 |
+
for pose_image in pose_images:
|
436 |
+
pose_cond_tensor = self.cond_image_processor.preprocess(
|
437 |
+
pose_image, height=height, width=width
|
438 |
+
)
|
439 |
+
pose_cond_tensor = pose_cond_tensor.unsqueeze(2) # (bs, c, 1, h, w)
|
440 |
+
pose_cond_tensor_list.append(pose_cond_tensor)
|
441 |
+
pose_cond_tensor = torch.cat(pose_cond_tensor_list, dim=2) # (bs, c, t, h, w)
|
442 |
+
|
443 |
+
pose_cond_tensor = pose_cond_tensor.to(
|
444 |
+
device=device, dtype=self.pose_guider.dtype
|
445 |
+
)
|
446 |
+
|
447 |
+
ref_pose_tensor = self.cond_image_processor.preprocess(
|
448 |
+
ref_pose_image, height=height, width=width
|
449 |
+
)
|
450 |
+
ref_pose_tensor = ref_pose_tensor.to(
|
451 |
+
device=device, dtype=self.pose_guider.dtype
|
452 |
+
)
|
453 |
+
|
454 |
+
context_scheduler = get_context_scheduler(context_schedule)
|
455 |
+
|
456 |
+
# denoising loop
|
457 |
+
num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
|
458 |
+
with self.progress_bar(total=num_inference_steps) as progress_bar:
|
459 |
+
for i, t in enumerate(timesteps):
|
460 |
+
noise_pred = torch.zeros(
|
461 |
+
(
|
462 |
+
latents.shape[0] * (2 if do_classifier_free_guidance else 1),
|
463 |
+
*latents.shape[1:],
|
464 |
+
),
|
465 |
+
device=latents.device,
|
466 |
+
dtype=latents.dtype,
|
467 |
+
)
|
468 |
+
counter = torch.zeros(
|
469 |
+
(1, 1, latents.shape[2], 1, 1),
|
470 |
+
device=latents.device,
|
471 |
+
dtype=latents.dtype,
|
472 |
+
)
|
473 |
+
|
474 |
+
# 1. Forward reference image
|
475 |
+
if i == 0:
|
476 |
+
self.reference_unet(
|
477 |
+
ref_image_latents.repeat(
|
478 |
+
(2 if do_classifier_free_guidance else 1), 1, 1, 1
|
479 |
+
),
|
480 |
+
torch.zeros_like(t),
|
481 |
+
# t,
|
482 |
+
encoder_hidden_states=encoder_hidden_states,
|
483 |
+
return_dict=False,
|
484 |
+
)
|
485 |
+
reference_control_reader.update(reference_control_writer)
|
486 |
+
|
487 |
+
context_queue = list(
|
488 |
+
context_scheduler(
|
489 |
+
0,
|
490 |
+
num_inference_steps,
|
491 |
+
latents.shape[2],
|
492 |
+
context_frames,
|
493 |
+
context_stride,
|
494 |
+
0,
|
495 |
+
)
|
496 |
+
)
|
497 |
+
num_context_batches = math.ceil(len(context_queue) / context_batch_size)
|
498 |
+
|
499 |
+
context_queue = list(
|
500 |
+
context_scheduler(
|
501 |
+
0,
|
502 |
+
num_inference_steps,
|
503 |
+
latents.shape[2],
|
504 |
+
context_frames,
|
505 |
+
context_stride,
|
506 |
+
context_overlap,
|
507 |
+
)
|
508 |
+
)
|
509 |
+
|
510 |
+
num_context_batches = math.ceil(len(context_queue) / context_batch_size)
|
511 |
+
global_context = []
|
512 |
+
for i in range(num_context_batches):
|
513 |
+
global_context.append(
|
514 |
+
context_queue[
|
515 |
+
i * context_batch_size : (i + 1) * context_batch_size
|
516 |
+
]
|
517 |
+
)
|
518 |
+
|
519 |
+
for context in global_context:
|
520 |
+
# 3.1 expand the latents if we are doing classifier free guidance
|
521 |
+
latent_model_input = (
|
522 |
+
torch.cat([latents[:, :, c] for c in context])
|
523 |
+
.to(device)
|
524 |
+
.repeat(2 if do_classifier_free_guidance else 1, 1, 1, 1, 1)
|
525 |
+
)
|
526 |
+
latent_model_input = self.scheduler.scale_model_input(
|
527 |
+
latent_model_input, t
|
528 |
+
)
|
529 |
+
b, c, f, h, w = latent_model_input.shape
|
530 |
+
|
531 |
+
pose_cond_input = (
|
532 |
+
torch.cat([pose_cond_tensor[:, :, c] for c in context])
|
533 |
+
.to(device)
|
534 |
+
.repeat(2 if do_classifier_free_guidance else 1, 1, 1, 1, 1)
|
535 |
+
)
|
536 |
+
pose_fea = self.pose_guider(pose_cond_input, ref_pose_tensor)
|
537 |
+
|
538 |
+
pred = self.denoising_unet(
|
539 |
+
latent_model_input,
|
540 |
+
t,
|
541 |
+
encoder_hidden_states=encoder_hidden_states[:b],
|
542 |
+
pose_cond_fea=pose_fea,
|
543 |
+
return_dict=False,
|
544 |
+
)[0]
|
545 |
+
|
546 |
+
for j, c in enumerate(context):
|
547 |
+
noise_pred[:, :, c] = noise_pred[:, :, c] + pred
|
548 |
+
counter[:, :, c] = counter[:, :, c] + 1
|
549 |
+
|
550 |
+
# perform guidance
|
551 |
+
if do_classifier_free_guidance:
|
552 |
+
noise_pred_uncond, noise_pred_text = (noise_pred / counter).chunk(2)
|
553 |
+
noise_pred = noise_pred_uncond + guidance_scale * (
|
554 |
+
noise_pred_text - noise_pred_uncond
|
555 |
+
)
|
556 |
+
|
557 |
+
latents = self.scheduler.step(
|
558 |
+
noise_pred, t, latents, **extra_step_kwargs
|
559 |
+
).prev_sample
|
560 |
+
|
561 |
+
if i == len(timesteps) - 1 or (
|
562 |
+
(i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0
|
563 |
+
):
|
564 |
+
progress_bar.update()
|
565 |
+
if callback is not None and i % callback_steps == 0:
|
566 |
+
step_idx = i // getattr(self.scheduler, "order", 1)
|
567 |
+
callback(step_idx, t, latents)
|
568 |
+
|
569 |
+
reference_control_reader.clear()
|
570 |
+
reference_control_writer.clear()
|
571 |
+
|
572 |
+
if interpolation_factor > 0:
|
573 |
+
latents = self.interpolate_latents(latents, interpolation_factor, device)
|
574 |
+
# Post-processing
|
575 |
+
images = self.decode_latents(latents) # (b, c, f, h, w)
|
576 |
+
|
577 |
+
# Convert to tensor
|
578 |
+
if output_type == "tensor":
|
579 |
+
images = torch.from_numpy(images)
|
580 |
+
|
581 |
+
if not return_dict:
|
582 |
+
return images
|
583 |
+
|
584 |
+
return Pose2VideoPipelineOutput(videos=images)
|
src/pipelines/utils.py
ADDED
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import torch
|
2 |
+
|
3 |
+
tensor_interpolation = None
|
4 |
+
|
5 |
+
|
6 |
+
def get_tensor_interpolation_method():
|
7 |
+
return tensor_interpolation
|
8 |
+
|
9 |
+
|
10 |
+
def set_tensor_interpolation_method(is_slerp):
|
11 |
+
global tensor_interpolation
|
12 |
+
tensor_interpolation = slerp if is_slerp else linear
|
13 |
+
|
14 |
+
|
15 |
+
def linear(v1, v2, t):
|
16 |
+
return (1.0 - t) * v1 + t * v2
|
17 |
+
|
18 |
+
|
19 |
+
def slerp(
|
20 |
+
v0: torch.Tensor, v1: torch.Tensor, t: float, DOT_THRESHOLD: float = 0.9995
|
21 |
+
) -> torch.Tensor:
|
22 |
+
u0 = v0 / v0.norm()
|
23 |
+
u1 = v1 / v1.norm()
|
24 |
+
dot = (u0 * u1).sum()
|
25 |
+
if dot.abs() > DOT_THRESHOLD:
|
26 |
+
# logger.info(f'warning: v0 and v1 close to parallel, using linear interpolation instead.')
|
27 |
+
return (1.0 - t) * v0 + t * v1
|
28 |
+
omega = dot.acos()
|
29 |
+
return (((1.0 - t) * omega).sin() * v0 + (t * omega).sin() * v1) / omega.sin()
|
src/utils/audio_util.py
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import math
|
3 |
+
|
4 |
+
import librosa
|
5 |
+
import numpy as np
|
6 |
+
from transformers import Wav2Vec2FeatureExtractor
|
7 |
+
|
8 |
+
|
9 |
+
class DataProcessor:
|
10 |
+
def __init__(self, sampling_rate, wav2vec_model_path):
|
11 |
+
self._processor = Wav2Vec2FeatureExtractor.from_pretrained(wav2vec_model_path, local_files_only=True)
|
12 |
+
self._sampling_rate = sampling_rate
|
13 |
+
|
14 |
+
def extract_feature(self, audio_path):
|
15 |
+
speech_array, sampling_rate = librosa.load(audio_path, sr=self._sampling_rate)
|
16 |
+
input_value = np.squeeze(self._processor(speech_array, sampling_rate=sampling_rate).input_values)
|
17 |
+
return input_value
|
18 |
+
|
19 |
+
|
20 |
+
def prepare_audio_feature(wav_file, fps=30, sampling_rate=16000, wav2vec_model_path=None):
|
21 |
+
data_preprocessor = DataProcessor(sampling_rate, wav2vec_model_path)
|
22 |
+
|
23 |
+
input_value = data_preprocessor.extract_feature(wav_file)
|
24 |
+
seq_len = math.ceil(len(input_value)/sampling_rate*fps)
|
25 |
+
return {
|
26 |
+
"audio_feature": input_value,
|
27 |
+
"seq_len": seq_len
|
28 |
+
}
|
29 |
+
|
30 |
+
|
src/utils/draw_util.py
ADDED
@@ -0,0 +1,149 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import cv2
|
2 |
+
import mediapipe as mp
|
3 |
+
import numpy as np
|
4 |
+
from mediapipe.framework.formats import landmark_pb2
|
5 |
+
|
6 |
+
class FaceMeshVisualizer:
|
7 |
+
def __init__(self, forehead_edge=False):
|
8 |
+
self.mp_drawing = mp.solutions.drawing_utils
|
9 |
+
mp_face_mesh = mp.solutions.face_mesh
|
10 |
+
self.mp_face_mesh = mp_face_mesh
|
11 |
+
self.forehead_edge = forehead_edge
|
12 |
+
|
13 |
+
DrawingSpec = mp.solutions.drawing_styles.DrawingSpec
|
14 |
+
f_thick = 2
|
15 |
+
f_rad = 1
|
16 |
+
right_iris_draw = DrawingSpec(color=(10, 200, 250), thickness=f_thick, circle_radius=f_rad)
|
17 |
+
right_eye_draw = DrawingSpec(color=(10, 200, 180), thickness=f_thick, circle_radius=f_rad)
|
18 |
+
right_eyebrow_draw = DrawingSpec(color=(10, 220, 180), thickness=f_thick, circle_radius=f_rad)
|
19 |
+
left_iris_draw = DrawingSpec(color=(250, 200, 10), thickness=f_thick, circle_radius=f_rad)
|
20 |
+
left_eye_draw = DrawingSpec(color=(180, 200, 10), thickness=f_thick, circle_radius=f_rad)
|
21 |
+
left_eyebrow_draw = DrawingSpec(color=(180, 220, 10), thickness=f_thick, circle_radius=f_rad)
|
22 |
+
head_draw = DrawingSpec(color=(10, 200, 10), thickness=f_thick, circle_radius=f_rad)
|
23 |
+
|
24 |
+
mouth_draw_obl = DrawingSpec(color=(10, 180, 20), thickness=f_thick, circle_radius=f_rad)
|
25 |
+
mouth_draw_obr = DrawingSpec(color=(20, 10, 180), thickness=f_thick, circle_radius=f_rad)
|
26 |
+
|
27 |
+
mouth_draw_ibl = DrawingSpec(color=(100, 100, 30), thickness=f_thick, circle_radius=f_rad)
|
28 |
+
mouth_draw_ibr = DrawingSpec(color=(100, 150, 50), thickness=f_thick, circle_radius=f_rad)
|
29 |
+
|
30 |
+
mouth_draw_otl = DrawingSpec(color=(20, 80, 100), thickness=f_thick, circle_radius=f_rad)
|
31 |
+
mouth_draw_otr = DrawingSpec(color=(80, 100, 20), thickness=f_thick, circle_radius=f_rad)
|
32 |
+
|
33 |
+
mouth_draw_itl = DrawingSpec(color=(120, 100, 200), thickness=f_thick, circle_radius=f_rad)
|
34 |
+
mouth_draw_itr = DrawingSpec(color=(150 ,120, 100), thickness=f_thick, circle_radius=f_rad)
|
35 |
+
|
36 |
+
FACEMESH_LIPS_OUTER_BOTTOM_LEFT = [(61,146),(146,91),(91,181),(181,84),(84,17)]
|
37 |
+
FACEMESH_LIPS_OUTER_BOTTOM_RIGHT = [(17,314),(314,405),(405,321),(321,375),(375,291)]
|
38 |
+
|
39 |
+
FACEMESH_LIPS_INNER_BOTTOM_LEFT = [(78,95),(95,88),(88,178),(178,87),(87,14)]
|
40 |
+
FACEMESH_LIPS_INNER_BOTTOM_RIGHT = [(14,317),(317,402),(402,318),(318,324),(324,308)]
|
41 |
+
|
42 |
+
FACEMESH_LIPS_OUTER_TOP_LEFT = [(61,185),(185,40),(40,39),(39,37),(37,0)]
|
43 |
+
FACEMESH_LIPS_OUTER_TOP_RIGHT = [(0,267),(267,269),(269,270),(270,409),(409,291)]
|
44 |
+
|
45 |
+
FACEMESH_LIPS_INNER_TOP_LEFT = [(78,191),(191,80),(80,81),(81,82),(82,13)]
|
46 |
+
FACEMESH_LIPS_INNER_TOP_RIGHT = [(13,312),(312,311),(311,310),(310,415),(415,308)]
|
47 |
+
|
48 |
+
FACEMESH_CUSTOM_FACE_OVAL = [(176, 149), (150, 136), (356, 454), (58, 132), (152, 148), (361, 288), (251, 389), (132, 93), (389, 356), (400, 377), (136, 172), (377, 152), (323, 361), (172, 58), (454, 323), (365, 379), (379, 378), (148, 176), (93, 234), (397, 365), (149, 150), (288, 397), (234, 127), (378, 400), (127, 162), (162, 21)]
|
49 |
+
|
50 |
+
# mp_face_mesh.FACEMESH_CONTOURS has all the items we care about.
|
51 |
+
face_connection_spec = {}
|
52 |
+
if self.forehead_edge:
|
53 |
+
for edge in mp_face_mesh.FACEMESH_FACE_OVAL:
|
54 |
+
face_connection_spec[edge] = head_draw
|
55 |
+
else:
|
56 |
+
for edge in FACEMESH_CUSTOM_FACE_OVAL:
|
57 |
+
face_connection_spec[edge] = head_draw
|
58 |
+
for edge in mp_face_mesh.FACEMESH_LEFT_EYE:
|
59 |
+
face_connection_spec[edge] = left_eye_draw
|
60 |
+
for edge in mp_face_mesh.FACEMESH_LEFT_EYEBROW:
|
61 |
+
face_connection_spec[edge] = left_eyebrow_draw
|
62 |
+
# for edge in mp_face_mesh.FACEMESH_LEFT_IRIS:
|
63 |
+
# face_connection_spec[edge] = left_iris_draw
|
64 |
+
for edge in mp_face_mesh.FACEMESH_RIGHT_EYE:
|
65 |
+
face_connection_spec[edge] = right_eye_draw
|
66 |
+
for edge in mp_face_mesh.FACEMESH_RIGHT_EYEBROW:
|
67 |
+
face_connection_spec[edge] = right_eyebrow_draw
|
68 |
+
# for edge in mp_face_mesh.FACEMESH_RIGHT_IRIS:
|
69 |
+
# face_connection_spec[edge] = right_iris_draw
|
70 |
+
# for edge in mp_face_mesh.FACEMESH_LIPS:
|
71 |
+
# face_connection_spec[edge] = mouth_draw
|
72 |
+
|
73 |
+
for edge in FACEMESH_LIPS_OUTER_BOTTOM_LEFT:
|
74 |
+
face_connection_spec[edge] = mouth_draw_obl
|
75 |
+
for edge in FACEMESH_LIPS_OUTER_BOTTOM_RIGHT:
|
76 |
+
face_connection_spec[edge] = mouth_draw_obr
|
77 |
+
for edge in FACEMESH_LIPS_INNER_BOTTOM_LEFT:
|
78 |
+
face_connection_spec[edge] = mouth_draw_ibl
|
79 |
+
for edge in FACEMESH_LIPS_INNER_BOTTOM_RIGHT:
|
80 |
+
face_connection_spec[edge] = mouth_draw_ibr
|
81 |
+
for edge in FACEMESH_LIPS_OUTER_TOP_LEFT:
|
82 |
+
face_connection_spec[edge] = mouth_draw_otl
|
83 |
+
for edge in FACEMESH_LIPS_OUTER_TOP_RIGHT:
|
84 |
+
face_connection_spec[edge] = mouth_draw_otr
|
85 |
+
for edge in FACEMESH_LIPS_INNER_TOP_LEFT:
|
86 |
+
face_connection_spec[edge] = mouth_draw_itl
|
87 |
+
for edge in FACEMESH_LIPS_INNER_TOP_RIGHT:
|
88 |
+
face_connection_spec[edge] = mouth_draw_itr
|
89 |
+
|
90 |
+
|
91 |
+
iris_landmark_spec = {468: right_iris_draw, 473: left_iris_draw}
|
92 |
+
|
93 |
+
self.face_connection_spec = face_connection_spec
|
94 |
+
def draw_pupils(self, image, landmark_list, drawing_spec, halfwidth: int = 2):
|
95 |
+
"""We have a custom function to draw the pupils because the mp.draw_landmarks method requires a parameter for all
|
96 |
+
landmarks. Until our PR is merged into mediapipe, we need this separate method."""
|
97 |
+
if len(image.shape) != 3:
|
98 |
+
raise ValueError("Input image must be H,W,C.")
|
99 |
+
image_rows, image_cols, image_channels = image.shape
|
100 |
+
if image_channels != 3: # BGR channels
|
101 |
+
raise ValueError('Input image must contain three channel bgr data.')
|
102 |
+
for idx, landmark in enumerate(landmark_list.landmark):
|
103 |
+
if (
|
104 |
+
(landmark.HasField('visibility') and landmark.visibility < 0.9) or
|
105 |
+
(landmark.HasField('presence') and landmark.presence < 0.5)
|
106 |
+
):
|
107 |
+
continue
|
108 |
+
if landmark.x >= 1.0 or landmark.x < 0 or landmark.y >= 1.0 or landmark.y < 0:
|
109 |
+
continue
|
110 |
+
image_x = int(image_cols*landmark.x)
|
111 |
+
image_y = int(image_rows*landmark.y)
|
112 |
+
draw_color = None
|
113 |
+
if isinstance(drawing_spec, Mapping):
|
114 |
+
if drawing_spec.get(idx) is None:
|
115 |
+
continue
|
116 |
+
else:
|
117 |
+
draw_color = drawing_spec[idx].color
|
118 |
+
elif isinstance(drawing_spec, DrawingSpec):
|
119 |
+
draw_color = drawing_spec.color
|
120 |
+
image[image_y-halfwidth:image_y+halfwidth, image_x-halfwidth:image_x+halfwidth, :] = draw_color
|
121 |
+
|
122 |
+
|
123 |
+
|
124 |
+
def draw_landmarks(self, image_size, keypoints, normed=False):
|
125 |
+
ini_size = [512, 512]
|
126 |
+
image = np.zeros([ini_size[1], ini_size[0], 3], dtype=np.uint8)
|
127 |
+
new_landmarks = landmark_pb2.NormalizedLandmarkList()
|
128 |
+
for i in range(keypoints.shape[0]):
|
129 |
+
landmark = new_landmarks.landmark.add()
|
130 |
+
if normed:
|
131 |
+
landmark.x = keypoints[i, 0]
|
132 |
+
landmark.y = keypoints[i, 1]
|
133 |
+
else:
|
134 |
+
landmark.x = keypoints[i, 0] / image_size[0]
|
135 |
+
landmark.y = keypoints[i, 1] / image_size[1]
|
136 |
+
landmark.z = 1.0
|
137 |
+
|
138 |
+
self.mp_drawing.draw_landmarks(
|
139 |
+
image=image,
|
140 |
+
landmark_list=new_landmarks,
|
141 |
+
connections=self.face_connection_spec.keys(),
|
142 |
+
landmark_drawing_spec=None,
|
143 |
+
connection_drawing_spec=self.face_connection_spec
|
144 |
+
)
|
145 |
+
# draw_pupils(image, face_landmarks, iris_landmark_spec, 2)
|
146 |
+
image = cv2.resize(image, (image_size[0], image_size[1]))
|
147 |
+
|
148 |
+
return image
|
149 |
+
|
src/utils/face_landmark.py
ADDED
@@ -0,0 +1,3305 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Copyright 2023 The MediaPipe Authors.
|
2 |
+
#
|
3 |
+
# Licensed under the Apache License, Version 2.0 (the "License");
|
4 |
+
# you may not use this file except in compliance with the License.
|
5 |
+
# You may obtain a copy of the License at
|
6 |
+
#
|
7 |
+
# http://www.apache.org/licenses/LICENSE-2.0
|
8 |
+
#
|
9 |
+
# Unless required by applicable law or agreed to in writing, software
|
10 |
+
# distributed under the License is distributed on an "AS IS" BASIS,
|
11 |
+
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
12 |
+
# See the License for the specific language governing permissions and
|
13 |
+
# limitations under the License.
|
14 |
+
"""MediaPipe face landmarker task."""
|
15 |
+
|
16 |
+
import dataclasses
|
17 |
+
import enum
|
18 |
+
from typing import Callable, Mapping, Optional, List
|
19 |
+
|
20 |
+
import numpy as np
|
21 |
+
|
22 |
+
from mediapipe.framework.formats import classification_pb2
|
23 |
+
from mediapipe.framework.formats import landmark_pb2
|
24 |
+
from mediapipe.framework.formats import matrix_data_pb2
|
25 |
+
from mediapipe.python import packet_creator
|
26 |
+
from mediapipe.python import packet_getter
|
27 |
+
from mediapipe.python._framework_bindings import image as image_module
|
28 |
+
from mediapipe.python._framework_bindings import packet as packet_module
|
29 |
+
# pylint: disable=unused-import
|
30 |
+
from mediapipe.tasks.cc.vision.face_geometry.proto import face_geometry_pb2
|
31 |
+
# pylint: enable=unused-import
|
32 |
+
from mediapipe.tasks.cc.vision.face_landmarker.proto import face_landmarker_graph_options_pb2
|
33 |
+
from mediapipe.tasks.python.components.containers import category as category_module
|
34 |
+
from mediapipe.tasks.python.components.containers import landmark as landmark_module
|
35 |
+
from mediapipe.tasks.python.core import base_options as base_options_module
|
36 |
+
from mediapipe.tasks.python.core import task_info as task_info_module
|
37 |
+
from mediapipe.tasks.python.core.optional_dependencies import doc_controls
|
38 |
+
from mediapipe.tasks.python.vision.core import base_vision_task_api
|
39 |
+
from mediapipe.tasks.python.vision.core import image_processing_options as image_processing_options_module
|
40 |
+
from mediapipe.tasks.python.vision.core import vision_task_running_mode as running_mode_module
|
41 |
+
|
42 |
+
_BaseOptions = base_options_module.BaseOptions
|
43 |
+
_FaceLandmarkerGraphOptionsProto = (
|
44 |
+
face_landmarker_graph_options_pb2.FaceLandmarkerGraphOptions
|
45 |
+
)
|
46 |
+
_LayoutEnum = matrix_data_pb2.MatrixData.Layout
|
47 |
+
_RunningMode = running_mode_module.VisionTaskRunningMode
|
48 |
+
_ImageProcessingOptions = image_processing_options_module.ImageProcessingOptions
|
49 |
+
_TaskInfo = task_info_module.TaskInfo
|
50 |
+
|
51 |
+
_IMAGE_IN_STREAM_NAME = 'image_in'
|
52 |
+
_IMAGE_OUT_STREAM_NAME = 'image_out'
|
53 |
+
_IMAGE_TAG = 'IMAGE'
|
54 |
+
_NORM_RECT_STREAM_NAME = 'norm_rect_in'
|
55 |
+
_NORM_RECT_TAG = 'NORM_RECT'
|
56 |
+
_NORM_LANDMARKS_STREAM_NAME = 'norm_landmarks'
|
57 |
+
_NORM_LANDMARKS_TAG = 'NORM_LANDMARKS'
|
58 |
+
_BLENDSHAPES_STREAM_NAME = 'blendshapes'
|
59 |
+
_BLENDSHAPES_TAG = 'BLENDSHAPES'
|
60 |
+
_FACE_GEOMETRY_STREAM_NAME = 'face_geometry'
|
61 |
+
_FACE_GEOMETRY_TAG = 'FACE_GEOMETRY'
|
62 |
+
_TASK_GRAPH_NAME = 'mediapipe.tasks.vision.face_landmarker.FaceLandmarkerGraph'
|
63 |
+
_MICRO_SECONDS_PER_MILLISECOND = 1000
|
64 |
+
|
65 |
+
|
66 |
+
class Blendshapes(enum.IntEnum):
|
67 |
+
"""The 52 blendshape coefficients."""
|
68 |
+
|
69 |
+
NEUTRAL = 0
|
70 |
+
BROW_DOWN_LEFT = 1
|
71 |
+
BROW_DOWN_RIGHT = 2
|
72 |
+
BROW_INNER_UP = 3
|
73 |
+
BROW_OUTER_UP_LEFT = 4
|
74 |
+
BROW_OUTER_UP_RIGHT = 5
|
75 |
+
CHEEK_PUFF = 6
|
76 |
+
CHEEK_SQUINT_LEFT = 7
|
77 |
+
CHEEK_SQUINT_RIGHT = 8
|
78 |
+
EYE_BLINK_LEFT = 9
|
79 |
+
EYE_BLINK_RIGHT = 10
|
80 |
+
EYE_LOOK_DOWN_LEFT = 11
|
81 |
+
EYE_LOOK_DOWN_RIGHT = 12
|
82 |
+
EYE_LOOK_IN_LEFT = 13
|
83 |
+
EYE_LOOK_IN_RIGHT = 14
|
84 |
+
EYE_LOOK_OUT_LEFT = 15
|
85 |
+
EYE_LOOK_OUT_RIGHT = 16
|
86 |
+
EYE_LOOK_UP_LEFT = 17
|
87 |
+
EYE_LOOK_UP_RIGHT = 18
|
88 |
+
EYE_SQUINT_LEFT = 19
|
89 |
+
EYE_SQUINT_RIGHT = 20
|
90 |
+
EYE_WIDE_LEFT = 21
|
91 |
+
EYE_WIDE_RIGHT = 22
|
92 |
+
JAW_FORWARD = 23
|
93 |
+
JAW_LEFT = 24
|
94 |
+
JAW_OPEN = 25
|
95 |
+
JAW_RIGHT = 26
|
96 |
+
MOUTH_CLOSE = 27
|
97 |
+
MOUTH_DIMPLE_LEFT = 28
|
98 |
+
MOUTH_DIMPLE_RIGHT = 29
|
99 |
+
MOUTH_FROWN_LEFT = 30
|
100 |
+
MOUTH_FROWN_RIGHT = 31
|
101 |
+
MOUTH_FUNNEL = 32
|
102 |
+
MOUTH_LEFT = 33
|
103 |
+
MOUTH_LOWER_DOWN_LEFT = 34
|
104 |
+
MOUTH_LOWER_DOWN_RIGHT = 35
|
105 |
+
MOUTH_PRESS_LEFT = 36
|
106 |
+
MOUTH_PRESS_RIGHT = 37
|
107 |
+
MOUTH_PUCKER = 38
|
108 |
+
MOUTH_RIGHT = 39
|
109 |
+
MOUTH_ROLL_LOWER = 40
|
110 |
+
MOUTH_ROLL_UPPER = 41
|
111 |
+
MOUTH_SHRUG_LOWER = 42
|
112 |
+
MOUTH_SHRUG_UPPER = 43
|
113 |
+
MOUTH_SMILE_LEFT = 44
|
114 |
+
MOUTH_SMILE_RIGHT = 45
|
115 |
+
MOUTH_STRETCH_LEFT = 46
|
116 |
+
MOUTH_STRETCH_RIGHT = 47
|
117 |
+
MOUTH_UPPER_UP_LEFT = 48
|
118 |
+
MOUTH_UPPER_UP_RIGHT = 49
|
119 |
+
NOSE_SNEER_LEFT = 50
|
120 |
+
NOSE_SNEER_RIGHT = 51
|
121 |
+
|
122 |
+
|
123 |
+
class FaceLandmarksConnections:
|
124 |
+
"""The connections between face landmarks."""
|
125 |
+
|
126 |
+
@dataclasses.dataclass
|
127 |
+
class Connection:
|
128 |
+
"""The connection class for face landmarks."""
|
129 |
+
|
130 |
+
start: int
|
131 |
+
end: int
|
132 |
+
|
133 |
+
FACE_LANDMARKS_LIPS: List[Connection] = [
|
134 |
+
Connection(61, 146),
|
135 |
+
Connection(146, 91),
|
136 |
+
Connection(91, 181),
|
137 |
+
Connection(181, 84),
|
138 |
+
Connection(84, 17),
|
139 |
+
Connection(17, 314),
|
140 |
+
Connection(314, 405),
|
141 |
+
Connection(405, 321),
|
142 |
+
Connection(321, 375),
|
143 |
+
Connection(375, 291),
|
144 |
+
Connection(61, 185),
|
145 |
+
Connection(185, 40),
|
146 |
+
Connection(40, 39),
|
147 |
+
Connection(39, 37),
|
148 |
+
Connection(37, 0),
|
149 |
+
Connection(0, 267),
|
150 |
+
Connection(267, 269),
|
151 |
+
Connection(269, 270),
|
152 |
+
Connection(270, 409),
|
153 |
+
Connection(409, 291),
|
154 |
+
Connection(78, 95),
|
155 |
+
Connection(95, 88),
|
156 |
+
Connection(88, 178),
|
157 |
+
Connection(178, 87),
|
158 |
+
Connection(87, 14),
|
159 |
+
Connection(14, 317),
|
160 |
+
Connection(317, 402),
|
161 |
+
Connection(402, 318),
|
162 |
+
Connection(318, 324),
|
163 |
+
Connection(324, 308),
|
164 |
+
Connection(78, 191),
|
165 |
+
Connection(191, 80),
|
166 |
+
Connection(80, 81),
|
167 |
+
Connection(81, 82),
|
168 |
+
Connection(82, 13),
|
169 |
+
Connection(13, 312),
|
170 |
+
Connection(312, 311),
|
171 |
+
Connection(311, 310),
|
172 |
+
Connection(310, 415),
|
173 |
+
Connection(415, 308),
|
174 |
+
]
|
175 |
+
|
176 |
+
FACE_LANDMARKS_LEFT_EYE: List[Connection] = [
|
177 |
+
Connection(263, 249),
|
178 |
+
Connection(249, 390),
|
179 |
+
Connection(390, 373),
|
180 |
+
Connection(373, 374),
|
181 |
+
Connection(374, 380),
|
182 |
+
Connection(380, 381),
|
183 |
+
Connection(381, 382),
|
184 |
+
Connection(382, 362),
|
185 |
+
Connection(263, 466),
|
186 |
+
Connection(466, 388),
|
187 |
+
Connection(388, 387),
|
188 |
+
Connection(387, 386),
|
189 |
+
Connection(386, 385),
|
190 |
+
Connection(385, 384),
|
191 |
+
Connection(384, 398),
|
192 |
+
Connection(398, 362),
|
193 |
+
]
|
194 |
+
|
195 |
+
FACE_LANDMARKS_LEFT_EYEBROW: List[Connection] = [
|
196 |
+
Connection(276, 283),
|
197 |
+
Connection(283, 282),
|
198 |
+
Connection(282, 295),
|
199 |
+
Connection(295, 285),
|
200 |
+
Connection(300, 293),
|
201 |
+
Connection(293, 334),
|
202 |
+
Connection(334, 296),
|
203 |
+
Connection(296, 336),
|
204 |
+
]
|
205 |
+
|
206 |
+
FACE_LANDMARKS_LEFT_IRIS: List[Connection] = [
|
207 |
+
Connection(474, 475),
|
208 |
+
Connection(475, 476),
|
209 |
+
Connection(476, 477),
|
210 |
+
Connection(477, 474),
|
211 |
+
]
|
212 |
+
|
213 |
+
FACE_LANDMARKS_RIGHT_EYE: List[Connection] = [
|
214 |
+
Connection(33, 7),
|
215 |
+
Connection(7, 163),
|
216 |
+
Connection(163, 144),
|
217 |
+
Connection(144, 145),
|
218 |
+
Connection(145, 153),
|
219 |
+
Connection(153, 154),
|
220 |
+
Connection(154, 155),
|
221 |
+
Connection(155, 133),
|
222 |
+
Connection(33, 246),
|
223 |
+
Connection(246, 161),
|
224 |
+
Connection(161, 160),
|
225 |
+
Connection(160, 159),
|
226 |
+
Connection(159, 158),
|
227 |
+
Connection(158, 157),
|
228 |
+
Connection(157, 173),
|
229 |
+
Connection(173, 133),
|
230 |
+
]
|
231 |
+
|
232 |
+
FACE_LANDMARKS_RIGHT_EYEBROW: List[Connection] = [
|
233 |
+
Connection(46, 53),
|
234 |
+
Connection(53, 52),
|
235 |
+
Connection(52, 65),
|
236 |
+
Connection(65, 55),
|
237 |
+
Connection(70, 63),
|
238 |
+
Connection(63, 105),
|
239 |
+
Connection(105, 66),
|
240 |
+
Connection(66, 107),
|
241 |
+
]
|
242 |
+
|
243 |
+
FACE_LANDMARKS_RIGHT_IRIS: List[Connection] = [
|
244 |
+
Connection(469, 470),
|
245 |
+
Connection(470, 471),
|
246 |
+
Connection(471, 472),
|
247 |
+
Connection(472, 469),
|
248 |
+
]
|
249 |
+
|
250 |
+
FACE_LANDMARKS_FACE_OVAL: List[Connection] = [
|
251 |
+
Connection(10, 338),
|
252 |
+
Connection(338, 297),
|
253 |
+
Connection(297, 332),
|
254 |
+
Connection(332, 284),
|
255 |
+
Connection(284, 251),
|
256 |
+
Connection(251, 389),
|
257 |
+
Connection(389, 356),
|
258 |
+
Connection(356, 454),
|
259 |
+
Connection(454, 323),
|
260 |
+
Connection(323, 361),
|
261 |
+
Connection(361, 288),
|
262 |
+
Connection(288, 397),
|
263 |
+
Connection(397, 365),
|
264 |
+
Connection(365, 379),
|
265 |
+
Connection(379, 378),
|
266 |
+
Connection(378, 400),
|
267 |
+
Connection(400, 377),
|
268 |
+
Connection(377, 152),
|
269 |
+
Connection(152, 148),
|
270 |
+
Connection(148, 176),
|
271 |
+
Connection(176, 149),
|
272 |
+
Connection(149, 150),
|
273 |
+
Connection(150, 136),
|
274 |
+
Connection(136, 172),
|
275 |
+
Connection(172, 58),
|
276 |
+
Connection(58, 132),
|
277 |
+
Connection(132, 93),
|
278 |
+
Connection(93, 234),
|
279 |
+
Connection(234, 127),
|
280 |
+
Connection(127, 162),
|
281 |
+
Connection(162, 21),
|
282 |
+
Connection(21, 54),
|
283 |
+
Connection(54, 103),
|
284 |
+
Connection(103, 67),
|
285 |
+
Connection(67, 109),
|
286 |
+
Connection(109, 10),
|
287 |
+
]
|
288 |
+
|
289 |
+
FACE_LANDMARKS_CONTOURS: List[Connection] = (
|
290 |
+
FACE_LANDMARKS_LIPS
|
291 |
+
+ FACE_LANDMARKS_LEFT_EYE
|
292 |
+
+ FACE_LANDMARKS_LEFT_EYEBROW
|
293 |
+
+ FACE_LANDMARKS_RIGHT_EYE
|
294 |
+
+ FACE_LANDMARKS_RIGHT_EYEBROW
|
295 |
+
+ FACE_LANDMARKS_FACE_OVAL
|
296 |
+
)
|
297 |
+
|
298 |
+
FACE_LANDMARKS_TESSELATION: List[Connection] = [
|
299 |
+
Connection(127, 34),
|
300 |
+
Connection(34, 139),
|
301 |
+
Connection(139, 127),
|
302 |
+
Connection(11, 0),
|
303 |
+
Connection(0, 37),
|
304 |
+
Connection(37, 11),
|
305 |
+
Connection(232, 231),
|
306 |
+
Connection(231, 120),
|
307 |
+
Connection(120, 232),
|
308 |
+
Connection(72, 37),
|
309 |
+
Connection(37, 39),
|
310 |
+
Connection(39, 72),
|
311 |
+
Connection(128, 121),
|
312 |
+
Connection(121, 47),
|
313 |
+
Connection(47, 128),
|
314 |
+
Connection(232, 121),
|
315 |
+
Connection(121, 128),
|
316 |
+
Connection(128, 232),
|
317 |
+
Connection(104, 69),
|
318 |
+
Connection(69, 67),
|
319 |
+
Connection(67, 104),
|
320 |
+
Connection(175, 171),
|
321 |
+
Connection(171, 148),
|
322 |
+
Connection(148, 175),
|
323 |
+
Connection(118, 50),
|
324 |
+
Connection(50, 101),
|
325 |
+
Connection(101, 118),
|
326 |
+
Connection(73, 39),
|
327 |
+
Connection(39, 40),
|
328 |
+
Connection(40, 73),
|
329 |
+
Connection(9, 151),
|
330 |
+
Connection(151, 108),
|
331 |
+
Connection(108, 9),
|
332 |
+
Connection(48, 115),
|
333 |
+
Connection(115, 131),
|
334 |
+
Connection(131, 48),
|
335 |
+
Connection(194, 204),
|
336 |
+
Connection(204, 211),
|
337 |
+
Connection(211, 194),
|
338 |
+
Connection(74, 40),
|
339 |
+
Connection(40, 185),
|
340 |
+
Connection(185, 74),
|
341 |
+
Connection(80, 42),
|
342 |
+
Connection(42, 183),
|
343 |
+
Connection(183, 80),
|
344 |
+
Connection(40, 92),
|
345 |
+
Connection(92, 186),
|
346 |
+
Connection(186, 40),
|
347 |
+
Connection(230, 229),
|
348 |
+
Connection(229, 118),
|
349 |
+
Connection(118, 230),
|
350 |
+
Connection(202, 212),
|
351 |
+
Connection(212, 214),
|
352 |
+
Connection(214, 202),
|
353 |
+
Connection(83, 18),
|
354 |
+
Connection(18, 17),
|
355 |
+
Connection(17, 83),
|
356 |
+
Connection(76, 61),
|
357 |
+
Connection(61, 146),
|
358 |
+
Connection(146, 76),
|
359 |
+
Connection(160, 29),
|
360 |
+
Connection(29, 30),
|
361 |
+
Connection(30, 160),
|
362 |
+
Connection(56, 157),
|
363 |
+
Connection(157, 173),
|
364 |
+
Connection(173, 56),
|
365 |
+
Connection(106, 204),
|
366 |
+
Connection(204, 194),
|
367 |
+
Connection(194, 106),
|
368 |
+
Connection(135, 214),
|
369 |
+
Connection(214, 192),
|
370 |
+
Connection(192, 135),
|
371 |
+
Connection(203, 165),
|
372 |
+
Connection(165, 98),
|
373 |
+
Connection(98, 203),
|
374 |
+
Connection(21, 71),
|
375 |
+
Connection(71, 68),
|
376 |
+
Connection(68, 21),
|
377 |
+
Connection(51, 45),
|
378 |
+
Connection(45, 4),
|
379 |
+
Connection(4, 51),
|
380 |
+
Connection(144, 24),
|
381 |
+
Connection(24, 23),
|
382 |
+
Connection(23, 144),
|
383 |
+
Connection(77, 146),
|
384 |
+
Connection(146, 91),
|
385 |
+
Connection(91, 77),
|
386 |
+
Connection(205, 50),
|
387 |
+
Connection(50, 187),
|
388 |
+
Connection(187, 205),
|
389 |
+
Connection(201, 200),
|
390 |
+
Connection(200, 18),
|
391 |
+
Connection(18, 201),
|
392 |
+
Connection(91, 106),
|
393 |
+
Connection(106, 182),
|
394 |
+
Connection(182, 91),
|
395 |
+
Connection(90, 91),
|
396 |
+
Connection(91, 181),
|
397 |
+
Connection(181, 90),
|
398 |
+
Connection(85, 84),
|
399 |
+
Connection(84, 17),
|
400 |
+
Connection(17, 85),
|
401 |
+
Connection(206, 203),
|
402 |
+
Connection(203, 36),
|
403 |
+
Connection(36, 206),
|
404 |
+
Connection(148, 171),
|
405 |
+
Connection(171, 140),
|
406 |
+
Connection(140, 148),
|
407 |
+
Connection(92, 40),
|
408 |
+
Connection(40, 39),
|
409 |
+
Connection(39, 92),
|
410 |
+
Connection(193, 189),
|
411 |
+
Connection(189, 244),
|
412 |
+
Connection(244, 193),
|
413 |
+
Connection(159, 158),
|
414 |
+
Connection(158, 28),
|
415 |
+
Connection(28, 159),
|
416 |
+
Connection(247, 246),
|
417 |
+
Connection(246, 161),
|
418 |
+
Connection(161, 247),
|
419 |
+
Connection(236, 3),
|
420 |
+
Connection(3, 196),
|
421 |
+
Connection(196, 236),
|
422 |
+
Connection(54, 68),
|
423 |
+
Connection(68, 104),
|
424 |
+
Connection(104, 54),
|
425 |
+
Connection(193, 168),
|
426 |
+
Connection(168, 8),
|
427 |
+
Connection(8, 193),
|
428 |
+
Connection(117, 228),
|
429 |
+
Connection(228, 31),
|
430 |
+
Connection(31, 117),
|
431 |
+
Connection(189, 193),
|
432 |
+
Connection(193, 55),
|
433 |
+
Connection(55, 189),
|
434 |
+
Connection(98, 97),
|
435 |
+
Connection(97, 99),
|
436 |
+
Connection(99, 98),
|
437 |
+
Connection(126, 47),
|
438 |
+
Connection(47, 100),
|
439 |
+
Connection(100, 126),
|
440 |
+
Connection(166, 79),
|
441 |
+
Connection(79, 218),
|
442 |
+
Connection(218, 166),
|
443 |
+
Connection(155, 154),
|
444 |
+
Connection(154, 26),
|
445 |
+
Connection(26, 155),
|
446 |
+
Connection(209, 49),
|
447 |
+
Connection(49, 131),
|
448 |
+
Connection(131, 209),
|
449 |
+
Connection(135, 136),
|
450 |
+
Connection(136, 150),
|
451 |
+
Connection(150, 135),
|
452 |
+
Connection(47, 126),
|
453 |
+
Connection(126, 217),
|
454 |
+
Connection(217, 47),
|
455 |
+
Connection(223, 52),
|
456 |
+
Connection(52, 53),
|
457 |
+
Connection(53, 223),
|
458 |
+
Connection(45, 51),
|
459 |
+
Connection(51, 134),
|
460 |
+
Connection(134, 45),
|
461 |
+
Connection(211, 170),
|
462 |
+
Connection(170, 140),
|
463 |
+
Connection(140, 211),
|
464 |
+
Connection(67, 69),
|
465 |
+
Connection(69, 108),
|
466 |
+
Connection(108, 67),
|
467 |
+
Connection(43, 106),
|
468 |
+
Connection(106, 91),
|
469 |
+
Connection(91, 43),
|
470 |
+
Connection(230, 119),
|
471 |
+
Connection(119, 120),
|
472 |
+
Connection(120, 230),
|
473 |
+
Connection(226, 130),
|
474 |
+
Connection(130, 247),
|
475 |
+
Connection(247, 226),
|
476 |
+
Connection(63, 53),
|
477 |
+
Connection(53, 52),
|
478 |
+
Connection(52, 63),
|
479 |
+
Connection(238, 20),
|
480 |
+
Connection(20, 242),
|
481 |
+
Connection(242, 238),
|
482 |
+
Connection(46, 70),
|
483 |
+
Connection(70, 156),
|
484 |
+
Connection(156, 46),
|
485 |
+
Connection(78, 62),
|
486 |
+
Connection(62, 96),
|
487 |
+
Connection(96, 78),
|
488 |
+
Connection(46, 53),
|
489 |
+
Connection(53, 63),
|
490 |
+
Connection(63, 46),
|
491 |
+
Connection(143, 34),
|
492 |
+
Connection(34, 227),
|
493 |
+
Connection(227, 143),
|
494 |
+
Connection(123, 117),
|
495 |
+
Connection(117, 111),
|
496 |
+
Connection(111, 123),
|
497 |
+
Connection(44, 125),
|
498 |
+
Connection(125, 19),
|
499 |
+
Connection(19, 44),
|
500 |
+
Connection(236, 134),
|
501 |
+
Connection(134, 51),
|
502 |
+
Connection(51, 236),
|
503 |
+
Connection(216, 206),
|
504 |
+
Connection(206, 205),
|
505 |
+
Connection(205, 216),
|
506 |
+
Connection(154, 153),
|
507 |
+
Connection(153, 22),
|
508 |
+
Connection(22, 154),
|
509 |
+
Connection(39, 37),
|
510 |
+
Connection(37, 167),
|
511 |
+
Connection(167, 39),
|
512 |
+
Connection(200, 201),
|
513 |
+
Connection(201, 208),
|
514 |
+
Connection(208, 200),
|
515 |
+
Connection(36, 142),
|
516 |
+
Connection(142, 100),
|
517 |
+
Connection(100, 36),
|
518 |
+
Connection(57, 212),
|
519 |
+
Connection(212, 202),
|
520 |
+
Connection(202, 57),
|
521 |
+
Connection(20, 60),
|
522 |
+
Connection(60, 99),
|
523 |
+
Connection(99, 20),
|
524 |
+
Connection(28, 158),
|
525 |
+
Connection(158, 157),
|
526 |
+
Connection(157, 28),
|
527 |
+
Connection(35, 226),
|
528 |
+
Connection(226, 113),
|
529 |
+
Connection(113, 35),
|
530 |
+
Connection(160, 159),
|
531 |
+
Connection(159, 27),
|
532 |
+
Connection(27, 160),
|
533 |
+
Connection(204, 202),
|
534 |
+
Connection(202, 210),
|
535 |
+
Connection(210, 204),
|
536 |
+
Connection(113, 225),
|
537 |
+
Connection(225, 46),
|
538 |
+
Connection(46, 113),
|
539 |
+
Connection(43, 202),
|
540 |
+
Connection(202, 204),
|
541 |
+
Connection(204, 43),
|
542 |
+
Connection(62, 76),
|
543 |
+
Connection(76, 77),
|
544 |
+
Connection(77, 62),
|
545 |
+
Connection(137, 123),
|
546 |
+
Connection(123, 116),
|
547 |
+
Connection(116, 137),
|
548 |
+
Connection(41, 38),
|
549 |
+
Connection(38, 72),
|
550 |
+
Connection(72, 41),
|
551 |
+
Connection(203, 129),
|
552 |
+
Connection(129, 142),
|
553 |
+
Connection(142, 203),
|
554 |
+
Connection(64, 98),
|
555 |
+
Connection(98, 240),
|
556 |
+
Connection(240, 64),
|
557 |
+
Connection(49, 102),
|
558 |
+
Connection(102, 64),
|
559 |
+
Connection(64, 49),
|
560 |
+
Connection(41, 73),
|
561 |
+
Connection(73, 74),
|
562 |
+
Connection(74, 41),
|
563 |
+
Connection(212, 216),
|
564 |
+
Connection(216, 207),
|
565 |
+
Connection(207, 212),
|
566 |
+
Connection(42, 74),
|
567 |
+
Connection(74, 184),
|
568 |
+
Connection(184, 42),
|
569 |
+
Connection(169, 170),
|
570 |
+
Connection(170, 211),
|
571 |
+
Connection(211, 169),
|
572 |
+
Connection(170, 149),
|
573 |
+
Connection(149, 176),
|
574 |
+
Connection(176, 170),
|
575 |
+
Connection(105, 66),
|
576 |
+
Connection(66, 69),
|
577 |
+
Connection(69, 105),
|
578 |
+
Connection(122, 6),
|
579 |
+
Connection(6, 168),
|
580 |
+
Connection(168, 122),
|
581 |
+
Connection(123, 147),
|
582 |
+
Connection(147, 187),
|
583 |
+
Connection(187, 123),
|
584 |
+
Connection(96, 77),
|
585 |
+
Connection(77, 90),
|
586 |
+
Connection(90, 96),
|
587 |
+
Connection(65, 55),
|
588 |
+
Connection(55, 107),
|
589 |
+
Connection(107, 65),
|
590 |
+
Connection(89, 90),
|
591 |
+
Connection(90, 180),
|
592 |
+
Connection(180, 89),
|
593 |
+
Connection(101, 100),
|
594 |
+
Connection(100, 120),
|
595 |
+
Connection(120, 101),
|
596 |
+
Connection(63, 105),
|
597 |
+
Connection(105, 104),
|
598 |
+
Connection(104, 63),
|
599 |
+
Connection(93, 137),
|
600 |
+
Connection(137, 227),
|
601 |
+
Connection(227, 93),
|
602 |
+
Connection(15, 86),
|
603 |
+
Connection(86, 85),
|
604 |
+
Connection(85, 15),
|
605 |
+
Connection(129, 102),
|
606 |
+
Connection(102, 49),
|
607 |
+
Connection(49, 129),
|
608 |
+
Connection(14, 87),
|
609 |
+
Connection(87, 86),
|
610 |
+
Connection(86, 14),
|
611 |
+
Connection(55, 8),
|
612 |
+
Connection(8, 9),
|
613 |
+
Connection(9, 55),
|
614 |
+
Connection(100, 47),
|
615 |
+
Connection(47, 121),
|
616 |
+
Connection(121, 100),
|
617 |
+
Connection(145, 23),
|
618 |
+
Connection(23, 22),
|
619 |
+
Connection(22, 145),
|
620 |
+
Connection(88, 89),
|
621 |
+
Connection(89, 179),
|
622 |
+
Connection(179, 88),
|
623 |
+
Connection(6, 122),
|
624 |
+
Connection(122, 196),
|
625 |
+
Connection(196, 6),
|
626 |
+
Connection(88, 95),
|
627 |
+
Connection(95, 96),
|
628 |
+
Connection(96, 88),
|
629 |
+
Connection(138, 172),
|
630 |
+
Connection(172, 136),
|
631 |
+
Connection(136, 138),
|
632 |
+
Connection(215, 58),
|
633 |
+
Connection(58, 172),
|
634 |
+
Connection(172, 215),
|
635 |
+
Connection(115, 48),
|
636 |
+
Connection(48, 219),
|
637 |
+
Connection(219, 115),
|
638 |
+
Connection(42, 80),
|
639 |
+
Connection(80, 81),
|
640 |
+
Connection(81, 42),
|
641 |
+
Connection(195, 3),
|
642 |
+
Connection(3, 51),
|
643 |
+
Connection(51, 195),
|
644 |
+
Connection(43, 146),
|
645 |
+
Connection(146, 61),
|
646 |
+
Connection(61, 43),
|
647 |
+
Connection(171, 175),
|
648 |
+
Connection(175, 199),
|
649 |
+
Connection(199, 171),
|
650 |
+
Connection(81, 82),
|
651 |
+
Connection(82, 38),
|
652 |
+
Connection(38, 81),
|
653 |
+
Connection(53, 46),
|
654 |
+
Connection(46, 225),
|
655 |
+
Connection(225, 53),
|
656 |
+
Connection(144, 163),
|
657 |
+
Connection(163, 110),
|
658 |
+
Connection(110, 144),
|
659 |
+
Connection(52, 65),
|
660 |
+
Connection(65, 66),
|
661 |
+
Connection(66, 52),
|
662 |
+
Connection(229, 228),
|
663 |
+
Connection(228, 117),
|
664 |
+
Connection(117, 229),
|
665 |
+
Connection(34, 127),
|
666 |
+
Connection(127, 234),
|
667 |
+
Connection(234, 34),
|
668 |
+
Connection(107, 108),
|
669 |
+
Connection(108, 69),
|
670 |
+
Connection(69, 107),
|
671 |
+
Connection(109, 108),
|
672 |
+
Connection(108, 151),
|
673 |
+
Connection(151, 109),
|
674 |
+
Connection(48, 64),
|
675 |
+
Connection(64, 235),
|
676 |
+
Connection(235, 48),
|
677 |
+
Connection(62, 78),
|
678 |
+
Connection(78, 191),
|
679 |
+
Connection(191, 62),
|
680 |
+
Connection(129, 209),
|
681 |
+
Connection(209, 126),
|
682 |
+
Connection(126, 129),
|
683 |
+
Connection(111, 35),
|
684 |
+
Connection(35, 143),
|
685 |
+
Connection(143, 111),
|
686 |
+
Connection(117, 123),
|
687 |
+
Connection(123, 50),
|
688 |
+
Connection(50, 117),
|
689 |
+
Connection(222, 65),
|
690 |
+
Connection(65, 52),
|
691 |
+
Connection(52, 222),
|
692 |
+
Connection(19, 125),
|
693 |
+
Connection(125, 141),
|
694 |
+
Connection(141, 19),
|
695 |
+
Connection(221, 55),
|
696 |
+
Connection(55, 65),
|
697 |
+
Connection(65, 221),
|
698 |
+
Connection(3, 195),
|
699 |
+
Connection(195, 197),
|
700 |
+
Connection(197, 3),
|
701 |
+
Connection(25, 7),
|
702 |
+
Connection(7, 33),
|
703 |
+
Connection(33, 25),
|
704 |
+
Connection(220, 237),
|
705 |
+
Connection(237, 44),
|
706 |
+
Connection(44, 220),
|
707 |
+
Connection(70, 71),
|
708 |
+
Connection(71, 139),
|
709 |
+
Connection(139, 70),
|
710 |
+
Connection(122, 193),
|
711 |
+
Connection(193, 245),
|
712 |
+
Connection(245, 122),
|
713 |
+
Connection(247, 130),
|
714 |
+
Connection(130, 33),
|
715 |
+
Connection(33, 247),
|
716 |
+
Connection(71, 21),
|
717 |
+
Connection(21, 162),
|
718 |
+
Connection(162, 71),
|
719 |
+
Connection(170, 169),
|
720 |
+
Connection(169, 150),
|
721 |
+
Connection(150, 170),
|
722 |
+
Connection(188, 174),
|
723 |
+
Connection(174, 196),
|
724 |
+
Connection(196, 188),
|
725 |
+
Connection(216, 186),
|
726 |
+
Connection(186, 92),
|
727 |
+
Connection(92, 216),
|
728 |
+
Connection(2, 97),
|
729 |
+
Connection(97, 167),
|
730 |
+
Connection(167, 2),
|
731 |
+
Connection(141, 125),
|
732 |
+
Connection(125, 241),
|
733 |
+
Connection(241, 141),
|
734 |
+
Connection(164, 167),
|
735 |
+
Connection(167, 37),
|
736 |
+
Connection(37, 164),
|
737 |
+
Connection(72, 38),
|
738 |
+
Connection(38, 12),
|
739 |
+
Connection(12, 72),
|
740 |
+
Connection(38, 82),
|
741 |
+
Connection(82, 13),
|
742 |
+
Connection(13, 38),
|
743 |
+
Connection(63, 68),
|
744 |
+
Connection(68, 71),
|
745 |
+
Connection(71, 63),
|
746 |
+
Connection(226, 35),
|
747 |
+
Connection(35, 111),
|
748 |
+
Connection(111, 226),
|
749 |
+
Connection(101, 50),
|
750 |
+
Connection(50, 205),
|
751 |
+
Connection(205, 101),
|
752 |
+
Connection(206, 92),
|
753 |
+
Connection(92, 165),
|
754 |
+
Connection(165, 206),
|
755 |
+
Connection(209, 198),
|
756 |
+
Connection(198, 217),
|
757 |
+
Connection(217, 209),
|
758 |
+
Connection(165, 167),
|
759 |
+
Connection(167, 97),
|
760 |
+
Connection(97, 165),
|
761 |
+
Connection(220, 115),
|
762 |
+
Connection(115, 218),
|
763 |
+
Connection(218, 220),
|
764 |
+
Connection(133, 112),
|
765 |
+
Connection(112, 243),
|
766 |
+
Connection(243, 133),
|
767 |
+
Connection(239, 238),
|
768 |
+
Connection(238, 241),
|
769 |
+
Connection(241, 239),
|
770 |
+
Connection(214, 135),
|
771 |
+
Connection(135, 169),
|
772 |
+
Connection(169, 214),
|
773 |
+
Connection(190, 173),
|
774 |
+
Connection(173, 133),
|
775 |
+
Connection(133, 190),
|
776 |
+
Connection(171, 208),
|
777 |
+
Connection(208, 32),
|
778 |
+
Connection(32, 171),
|
779 |
+
Connection(125, 44),
|
780 |
+
Connection(44, 237),
|
781 |
+
Connection(237, 125),
|
782 |
+
Connection(86, 87),
|
783 |
+
Connection(87, 178),
|
784 |
+
Connection(178, 86),
|
785 |
+
Connection(85, 86),
|
786 |
+
Connection(86, 179),
|
787 |
+
Connection(179, 85),
|
788 |
+
Connection(84, 85),
|
789 |
+
Connection(85, 180),
|
790 |
+
Connection(180, 84),
|
791 |
+
Connection(83, 84),
|
792 |
+
Connection(84, 181),
|
793 |
+
Connection(181, 83),
|
794 |
+
Connection(201, 83),
|
795 |
+
Connection(83, 182),
|
796 |
+
Connection(182, 201),
|
797 |
+
Connection(137, 93),
|
798 |
+
Connection(93, 132),
|
799 |
+
Connection(132, 137),
|
800 |
+
Connection(76, 62),
|
801 |
+
Connection(62, 183),
|
802 |
+
Connection(183, 76),
|
803 |
+
Connection(61, 76),
|
804 |
+
Connection(76, 184),
|
805 |
+
Connection(184, 61),
|
806 |
+
Connection(57, 61),
|
807 |
+
Connection(61, 185),
|
808 |
+
Connection(185, 57),
|
809 |
+
Connection(212, 57),
|
810 |
+
Connection(57, 186),
|
811 |
+
Connection(186, 212),
|
812 |
+
Connection(214, 207),
|
813 |
+
Connection(207, 187),
|
814 |
+
Connection(187, 214),
|
815 |
+
Connection(34, 143),
|
816 |
+
Connection(143, 156),
|
817 |
+
Connection(156, 34),
|
818 |
+
Connection(79, 239),
|
819 |
+
Connection(239, 237),
|
820 |
+
Connection(237, 79),
|
821 |
+
Connection(123, 137),
|
822 |
+
Connection(137, 177),
|
823 |
+
Connection(177, 123),
|
824 |
+
Connection(44, 1),
|
825 |
+
Connection(1, 4),
|
826 |
+
Connection(4, 44),
|
827 |
+
Connection(201, 194),
|
828 |
+
Connection(194, 32),
|
829 |
+
Connection(32, 201),
|
830 |
+
Connection(64, 102),
|
831 |
+
Connection(102, 129),
|
832 |
+
Connection(129, 64),
|
833 |
+
Connection(213, 215),
|
834 |
+
Connection(215, 138),
|
835 |
+
Connection(138, 213),
|
836 |
+
Connection(59, 166),
|
837 |
+
Connection(166, 219),
|
838 |
+
Connection(219, 59),
|
839 |
+
Connection(242, 99),
|
840 |
+
Connection(99, 97),
|
841 |
+
Connection(97, 242),
|
842 |
+
Connection(2, 94),
|
843 |
+
Connection(94, 141),
|
844 |
+
Connection(141, 2),
|
845 |
+
Connection(75, 59),
|
846 |
+
Connection(59, 235),
|
847 |
+
Connection(235, 75),
|
848 |
+
Connection(24, 110),
|
849 |
+
Connection(110, 228),
|
850 |
+
Connection(228, 24),
|
851 |
+
Connection(25, 130),
|
852 |
+
Connection(130, 226),
|
853 |
+
Connection(226, 25),
|
854 |
+
Connection(23, 24),
|
855 |
+
Connection(24, 229),
|
856 |
+
Connection(229, 23),
|
857 |
+
Connection(22, 23),
|
858 |
+
Connection(23, 230),
|
859 |
+
Connection(230, 22),
|
860 |
+
Connection(26, 22),
|
861 |
+
Connection(22, 231),
|
862 |
+
Connection(231, 26),
|
863 |
+
Connection(112, 26),
|
864 |
+
Connection(26, 232),
|
865 |
+
Connection(232, 112),
|
866 |
+
Connection(189, 190),
|
867 |
+
Connection(190, 243),
|
868 |
+
Connection(243, 189),
|
869 |
+
Connection(221, 56),
|
870 |
+
Connection(56, 190),
|
871 |
+
Connection(190, 221),
|
872 |
+
Connection(28, 56),
|
873 |
+
Connection(56, 221),
|
874 |
+
Connection(221, 28),
|
875 |
+
Connection(27, 28),
|
876 |
+
Connection(28, 222),
|
877 |
+
Connection(222, 27),
|
878 |
+
Connection(29, 27),
|
879 |
+
Connection(27, 223),
|
880 |
+
Connection(223, 29),
|
881 |
+
Connection(30, 29),
|
882 |
+
Connection(29, 224),
|
883 |
+
Connection(224, 30),
|
884 |
+
Connection(247, 30),
|
885 |
+
Connection(30, 225),
|
886 |
+
Connection(225, 247),
|
887 |
+
Connection(238, 79),
|
888 |
+
Connection(79, 20),
|
889 |
+
Connection(20, 238),
|
890 |
+
Connection(166, 59),
|
891 |
+
Connection(59, 75),
|
892 |
+
Connection(75, 166),
|
893 |
+
Connection(60, 75),
|
894 |
+
Connection(75, 240),
|
895 |
+
Connection(240, 60),
|
896 |
+
Connection(147, 177),
|
897 |
+
Connection(177, 215),
|
898 |
+
Connection(215, 147),
|
899 |
+
Connection(20, 79),
|
900 |
+
Connection(79, 166),
|
901 |
+
Connection(166, 20),
|
902 |
+
Connection(187, 147),
|
903 |
+
Connection(147, 213),
|
904 |
+
Connection(213, 187),
|
905 |
+
Connection(112, 233),
|
906 |
+
Connection(233, 244),
|
907 |
+
Connection(244, 112),
|
908 |
+
Connection(233, 128),
|
909 |
+
Connection(128, 245),
|
910 |
+
Connection(245, 233),
|
911 |
+
Connection(128, 114),
|
912 |
+
Connection(114, 188),
|
913 |
+
Connection(188, 128),
|
914 |
+
Connection(114, 217),
|
915 |
+
Connection(217, 174),
|
916 |
+
Connection(174, 114),
|
917 |
+
Connection(131, 115),
|
918 |
+
Connection(115, 220),
|
919 |
+
Connection(220, 131),
|
920 |
+
Connection(217, 198),
|
921 |
+
Connection(198, 236),
|
922 |
+
Connection(236, 217),
|
923 |
+
Connection(198, 131),
|
924 |
+
Connection(131, 134),
|
925 |
+
Connection(134, 198),
|
926 |
+
Connection(177, 132),
|
927 |
+
Connection(132, 58),
|
928 |
+
Connection(58, 177),
|
929 |
+
Connection(143, 35),
|
930 |
+
Connection(35, 124),
|
931 |
+
Connection(124, 143),
|
932 |
+
Connection(110, 163),
|
933 |
+
Connection(163, 7),
|
934 |
+
Connection(7, 110),
|
935 |
+
Connection(228, 110),
|
936 |
+
Connection(110, 25),
|
937 |
+
Connection(25, 228),
|
938 |
+
Connection(356, 389),
|
939 |
+
Connection(389, 368),
|
940 |
+
Connection(368, 356),
|
941 |
+
Connection(11, 302),
|
942 |
+
Connection(302, 267),
|
943 |
+
Connection(267, 11),
|
944 |
+
Connection(452, 350),
|
945 |
+
Connection(350, 349),
|
946 |
+
Connection(349, 452),
|
947 |
+
Connection(302, 303),
|
948 |
+
Connection(303, 269),
|
949 |
+
Connection(269, 302),
|
950 |
+
Connection(357, 343),
|
951 |
+
Connection(343, 277),
|
952 |
+
Connection(277, 357),
|
953 |
+
Connection(452, 453),
|
954 |
+
Connection(453, 357),
|
955 |
+
Connection(357, 452),
|
956 |
+
Connection(333, 332),
|
957 |
+
Connection(332, 297),
|
958 |
+
Connection(297, 333),
|
959 |
+
Connection(175, 152),
|
960 |
+
Connection(152, 377),
|
961 |
+
Connection(377, 175),
|
962 |
+
Connection(347, 348),
|
963 |
+
Connection(348, 330),
|
964 |
+
Connection(330, 347),
|
965 |
+
Connection(303, 304),
|
966 |
+
Connection(304, 270),
|
967 |
+
Connection(270, 303),
|
968 |
+
Connection(9, 336),
|
969 |
+
Connection(336, 337),
|
970 |
+
Connection(337, 9),
|
971 |
+
Connection(278, 279),
|
972 |
+
Connection(279, 360),
|
973 |
+
Connection(360, 278),
|
974 |
+
Connection(418, 262),
|
975 |
+
Connection(262, 431),
|
976 |
+
Connection(431, 418),
|
977 |
+
Connection(304, 408),
|
978 |
+
Connection(408, 409),
|
979 |
+
Connection(409, 304),
|
980 |
+
Connection(310, 415),
|
981 |
+
Connection(415, 407),
|
982 |
+
Connection(407, 310),
|
983 |
+
Connection(270, 409),
|
984 |
+
Connection(409, 410),
|
985 |
+
Connection(410, 270),
|
986 |
+
Connection(450, 348),
|
987 |
+
Connection(348, 347),
|
988 |
+
Connection(347, 450),
|
989 |
+
Connection(422, 430),
|
990 |
+
Connection(430, 434),
|
991 |
+
Connection(434, 422),
|
992 |
+
Connection(313, 314),
|
993 |
+
Connection(314, 17),
|
994 |
+
Connection(17, 313),
|
995 |
+
Connection(306, 307),
|
996 |
+
Connection(307, 375),
|
997 |
+
Connection(375, 306),
|
998 |
+
Connection(387, 388),
|
999 |
+
Connection(388, 260),
|
1000 |
+
Connection(260, 387),
|
1001 |
+
Connection(286, 414),
|
1002 |
+
Connection(414, 398),
|
1003 |
+
Connection(398, 286),
|
1004 |
+
Connection(335, 406),
|
1005 |
+
Connection(406, 418),
|
1006 |
+
Connection(418, 335),
|
1007 |
+
Connection(364, 367),
|
1008 |
+
Connection(367, 416),
|
1009 |
+
Connection(416, 364),
|
1010 |
+
Connection(423, 358),
|
1011 |
+
Connection(358, 327),
|
1012 |
+
Connection(327, 423),
|
1013 |
+
Connection(251, 284),
|
1014 |
+
Connection(284, 298),
|
1015 |
+
Connection(298, 251),
|
1016 |
+
Connection(281, 5),
|
1017 |
+
Connection(5, 4),
|
1018 |
+
Connection(4, 281),
|
1019 |
+
Connection(373, 374),
|
1020 |
+
Connection(374, 253),
|
1021 |
+
Connection(253, 373),
|
1022 |
+
Connection(307, 320),
|
1023 |
+
Connection(320, 321),
|
1024 |
+
Connection(321, 307),
|
1025 |
+
Connection(425, 427),
|
1026 |
+
Connection(427, 411),
|
1027 |
+
Connection(411, 425),
|
1028 |
+
Connection(421, 313),
|
1029 |
+
Connection(313, 18),
|
1030 |
+
Connection(18, 421),
|
1031 |
+
Connection(321, 405),
|
1032 |
+
Connection(405, 406),
|
1033 |
+
Connection(406, 321),
|
1034 |
+
Connection(320, 404),
|
1035 |
+
Connection(404, 405),
|
1036 |
+
Connection(405, 320),
|
1037 |
+
Connection(315, 16),
|
1038 |
+
Connection(16, 17),
|
1039 |
+
Connection(17, 315),
|
1040 |
+
Connection(426, 425),
|
1041 |
+
Connection(425, 266),
|
1042 |
+
Connection(266, 426),
|
1043 |
+
Connection(377, 400),
|
1044 |
+
Connection(400, 369),
|
1045 |
+
Connection(369, 377),
|
1046 |
+
Connection(322, 391),
|
1047 |
+
Connection(391, 269),
|
1048 |
+
Connection(269, 322),
|
1049 |
+
Connection(417, 465),
|
1050 |
+
Connection(465, 464),
|
1051 |
+
Connection(464, 417),
|
1052 |
+
Connection(386, 257),
|
1053 |
+
Connection(257, 258),
|
1054 |
+
Connection(258, 386),
|
1055 |
+
Connection(466, 260),
|
1056 |
+
Connection(260, 388),
|
1057 |
+
Connection(388, 466),
|
1058 |
+
Connection(456, 399),
|
1059 |
+
Connection(399, 419),
|
1060 |
+
Connection(419, 456),
|
1061 |
+
Connection(284, 332),
|
1062 |
+
Connection(332, 333),
|
1063 |
+
Connection(333, 284),
|
1064 |
+
Connection(417, 285),
|
1065 |
+
Connection(285, 8),
|
1066 |
+
Connection(8, 417),
|
1067 |
+
Connection(346, 340),
|
1068 |
+
Connection(340, 261),
|
1069 |
+
Connection(261, 346),
|
1070 |
+
Connection(413, 441),
|
1071 |
+
Connection(441, 285),
|
1072 |
+
Connection(285, 413),
|
1073 |
+
Connection(327, 460),
|
1074 |
+
Connection(460, 328),
|
1075 |
+
Connection(328, 327),
|
1076 |
+
Connection(355, 371),
|
1077 |
+
Connection(371, 329),
|
1078 |
+
Connection(329, 355),
|
1079 |
+
Connection(392, 439),
|
1080 |
+
Connection(439, 438),
|
1081 |
+
Connection(438, 392),
|
1082 |
+
Connection(382, 341),
|
1083 |
+
Connection(341, 256),
|
1084 |
+
Connection(256, 382),
|
1085 |
+
Connection(429, 420),
|
1086 |
+
Connection(420, 360),
|
1087 |
+
Connection(360, 429),
|
1088 |
+
Connection(364, 394),
|
1089 |
+
Connection(394, 379),
|
1090 |
+
Connection(379, 364),
|
1091 |
+
Connection(277, 343),
|
1092 |
+
Connection(343, 437),
|
1093 |
+
Connection(437, 277),
|
1094 |
+
Connection(443, 444),
|
1095 |
+
Connection(444, 283),
|
1096 |
+
Connection(283, 443),
|
1097 |
+
Connection(275, 440),
|
1098 |
+
Connection(440, 363),
|
1099 |
+
Connection(363, 275),
|
1100 |
+
Connection(431, 262),
|
1101 |
+
Connection(262, 369),
|
1102 |
+
Connection(369, 431),
|
1103 |
+
Connection(297, 338),
|
1104 |
+
Connection(338, 337),
|
1105 |
+
Connection(337, 297),
|
1106 |
+
Connection(273, 375),
|
1107 |
+
Connection(375, 321),
|
1108 |
+
Connection(321, 273),
|
1109 |
+
Connection(450, 451),
|
1110 |
+
Connection(451, 349),
|
1111 |
+
Connection(349, 450),
|
1112 |
+
Connection(446, 342),
|
1113 |
+
Connection(342, 467),
|
1114 |
+
Connection(467, 446),
|
1115 |
+
Connection(293, 334),
|
1116 |
+
Connection(334, 282),
|
1117 |
+
Connection(282, 293),
|
1118 |
+
Connection(458, 461),
|
1119 |
+
Connection(461, 462),
|
1120 |
+
Connection(462, 458),
|
1121 |
+
Connection(276, 353),
|
1122 |
+
Connection(353, 383),
|
1123 |
+
Connection(383, 276),
|
1124 |
+
Connection(308, 324),
|
1125 |
+
Connection(324, 325),
|
1126 |
+
Connection(325, 308),
|
1127 |
+
Connection(276, 300),
|
1128 |
+
Connection(300, 293),
|
1129 |
+
Connection(293, 276),
|
1130 |
+
Connection(372, 345),
|
1131 |
+
Connection(345, 447),
|
1132 |
+
Connection(447, 372),
|
1133 |
+
Connection(352, 345),
|
1134 |
+
Connection(345, 340),
|
1135 |
+
Connection(340, 352),
|
1136 |
+
Connection(274, 1),
|
1137 |
+
Connection(1, 19),
|
1138 |
+
Connection(19, 274),
|
1139 |
+
Connection(456, 248),
|
1140 |
+
Connection(248, 281),
|
1141 |
+
Connection(281, 456),
|
1142 |
+
Connection(436, 427),
|
1143 |
+
Connection(427, 425),
|
1144 |
+
Connection(425, 436),
|
1145 |
+
Connection(381, 256),
|
1146 |
+
Connection(256, 252),
|
1147 |
+
Connection(252, 381),
|
1148 |
+
Connection(269, 391),
|
1149 |
+
Connection(391, 393),
|
1150 |
+
Connection(393, 269),
|
1151 |
+
Connection(200, 199),
|
1152 |
+
Connection(199, 428),
|
1153 |
+
Connection(428, 200),
|
1154 |
+
Connection(266, 330),
|
1155 |
+
Connection(330, 329),
|
1156 |
+
Connection(329, 266),
|
1157 |
+
Connection(287, 273),
|
1158 |
+
Connection(273, 422),
|
1159 |
+
Connection(422, 287),
|
1160 |
+
Connection(250, 462),
|
1161 |
+
Connection(462, 328),
|
1162 |
+
Connection(328, 250),
|
1163 |
+
Connection(258, 286),
|
1164 |
+
Connection(286, 384),
|
1165 |
+
Connection(384, 258),
|
1166 |
+
Connection(265, 353),
|
1167 |
+
Connection(353, 342),
|
1168 |
+
Connection(342, 265),
|
1169 |
+
Connection(387, 259),
|
1170 |
+
Connection(259, 257),
|
1171 |
+
Connection(257, 387),
|
1172 |
+
Connection(424, 431),
|
1173 |
+
Connection(431, 430),
|
1174 |
+
Connection(430, 424),
|
1175 |
+
Connection(342, 353),
|
1176 |
+
Connection(353, 276),
|
1177 |
+
Connection(276, 342),
|
1178 |
+
Connection(273, 335),
|
1179 |
+
Connection(335, 424),
|
1180 |
+
Connection(424, 273),
|
1181 |
+
Connection(292, 325),
|
1182 |
+
Connection(325, 307),
|
1183 |
+
Connection(307, 292),
|
1184 |
+
Connection(366, 447),
|
1185 |
+
Connection(447, 345),
|
1186 |
+
Connection(345, 366),
|
1187 |
+
Connection(271, 303),
|
1188 |
+
Connection(303, 302),
|
1189 |
+
Connection(302, 271),
|
1190 |
+
Connection(423, 266),
|
1191 |
+
Connection(266, 371),
|
1192 |
+
Connection(371, 423),
|
1193 |
+
Connection(294, 455),
|
1194 |
+
Connection(455, 460),
|
1195 |
+
Connection(460, 294),
|
1196 |
+
Connection(279, 278),
|
1197 |
+
Connection(278, 294),
|
1198 |
+
Connection(294, 279),
|
1199 |
+
Connection(271, 272),
|
1200 |
+
Connection(272, 304),
|
1201 |
+
Connection(304, 271),
|
1202 |
+
Connection(432, 434),
|
1203 |
+
Connection(434, 427),
|
1204 |
+
Connection(427, 432),
|
1205 |
+
Connection(272, 407),
|
1206 |
+
Connection(407, 408),
|
1207 |
+
Connection(408, 272),
|
1208 |
+
Connection(394, 430),
|
1209 |
+
Connection(430, 431),
|
1210 |
+
Connection(431, 394),
|
1211 |
+
Connection(395, 369),
|
1212 |
+
Connection(369, 400),
|
1213 |
+
Connection(400, 395),
|
1214 |
+
Connection(334, 333),
|
1215 |
+
Connection(333, 299),
|
1216 |
+
Connection(299, 334),
|
1217 |
+
Connection(351, 417),
|
1218 |
+
Connection(417, 168),
|
1219 |
+
Connection(168, 351),
|
1220 |
+
Connection(352, 280),
|
1221 |
+
Connection(280, 411),
|
1222 |
+
Connection(411, 352),
|
1223 |
+
Connection(325, 319),
|
1224 |
+
Connection(319, 320),
|
1225 |
+
Connection(320, 325),
|
1226 |
+
Connection(295, 296),
|
1227 |
+
Connection(296, 336),
|
1228 |
+
Connection(336, 295),
|
1229 |
+
Connection(319, 403),
|
1230 |
+
Connection(403, 404),
|
1231 |
+
Connection(404, 319),
|
1232 |
+
Connection(330, 348),
|
1233 |
+
Connection(348, 349),
|
1234 |
+
Connection(349, 330),
|
1235 |
+
Connection(293, 298),
|
1236 |
+
Connection(298, 333),
|
1237 |
+
Connection(333, 293),
|
1238 |
+
Connection(323, 454),
|
1239 |
+
Connection(454, 447),
|
1240 |
+
Connection(447, 323),
|
1241 |
+
Connection(15, 16),
|
1242 |
+
Connection(16, 315),
|
1243 |
+
Connection(315, 15),
|
1244 |
+
Connection(358, 429),
|
1245 |
+
Connection(429, 279),
|
1246 |
+
Connection(279, 358),
|
1247 |
+
Connection(14, 15),
|
1248 |
+
Connection(15, 316),
|
1249 |
+
Connection(316, 14),
|
1250 |
+
Connection(285, 336),
|
1251 |
+
Connection(336, 9),
|
1252 |
+
Connection(9, 285),
|
1253 |
+
Connection(329, 349),
|
1254 |
+
Connection(349, 350),
|
1255 |
+
Connection(350, 329),
|
1256 |
+
Connection(374, 380),
|
1257 |
+
Connection(380, 252),
|
1258 |
+
Connection(252, 374),
|
1259 |
+
Connection(318, 402),
|
1260 |
+
Connection(402, 403),
|
1261 |
+
Connection(403, 318),
|
1262 |
+
Connection(6, 197),
|
1263 |
+
Connection(197, 419),
|
1264 |
+
Connection(419, 6),
|
1265 |
+
Connection(318, 319),
|
1266 |
+
Connection(319, 325),
|
1267 |
+
Connection(325, 318),
|
1268 |
+
Connection(367, 364),
|
1269 |
+
Connection(364, 365),
|
1270 |
+
Connection(365, 367),
|
1271 |
+
Connection(435, 367),
|
1272 |
+
Connection(367, 397),
|
1273 |
+
Connection(397, 435),
|
1274 |
+
Connection(344, 438),
|
1275 |
+
Connection(438, 439),
|
1276 |
+
Connection(439, 344),
|
1277 |
+
Connection(272, 271),
|
1278 |
+
Connection(271, 311),
|
1279 |
+
Connection(311, 272),
|
1280 |
+
Connection(195, 5),
|
1281 |
+
Connection(5, 281),
|
1282 |
+
Connection(281, 195),
|
1283 |
+
Connection(273, 287),
|
1284 |
+
Connection(287, 291),
|
1285 |
+
Connection(291, 273),
|
1286 |
+
Connection(396, 428),
|
1287 |
+
Connection(428, 199),
|
1288 |
+
Connection(199, 396),
|
1289 |
+
Connection(311, 271),
|
1290 |
+
Connection(271, 268),
|
1291 |
+
Connection(268, 311),
|
1292 |
+
Connection(283, 444),
|
1293 |
+
Connection(444, 445),
|
1294 |
+
Connection(445, 283),
|
1295 |
+
Connection(373, 254),
|
1296 |
+
Connection(254, 339),
|
1297 |
+
Connection(339, 373),
|
1298 |
+
Connection(282, 334),
|
1299 |
+
Connection(334, 296),
|
1300 |
+
Connection(296, 282),
|
1301 |
+
Connection(449, 347),
|
1302 |
+
Connection(347, 346),
|
1303 |
+
Connection(346, 449),
|
1304 |
+
Connection(264, 447),
|
1305 |
+
Connection(447, 454),
|
1306 |
+
Connection(454, 264),
|
1307 |
+
Connection(336, 296),
|
1308 |
+
Connection(296, 299),
|
1309 |
+
Connection(299, 336),
|
1310 |
+
Connection(338, 10),
|
1311 |
+
Connection(10, 151),
|
1312 |
+
Connection(151, 338),
|
1313 |
+
Connection(278, 439),
|
1314 |
+
Connection(439, 455),
|
1315 |
+
Connection(455, 278),
|
1316 |
+
Connection(292, 407),
|
1317 |
+
Connection(407, 415),
|
1318 |
+
Connection(415, 292),
|
1319 |
+
Connection(358, 371),
|
1320 |
+
Connection(371, 355),
|
1321 |
+
Connection(355, 358),
|
1322 |
+
Connection(340, 345),
|
1323 |
+
Connection(345, 372),
|
1324 |
+
Connection(372, 340),
|
1325 |
+
Connection(346, 347),
|
1326 |
+
Connection(347, 280),
|
1327 |
+
Connection(280, 346),
|
1328 |
+
Connection(442, 443),
|
1329 |
+
Connection(443, 282),
|
1330 |
+
Connection(282, 442),
|
1331 |
+
Connection(19, 94),
|
1332 |
+
Connection(94, 370),
|
1333 |
+
Connection(370, 19),
|
1334 |
+
Connection(441, 442),
|
1335 |
+
Connection(442, 295),
|
1336 |
+
Connection(295, 441),
|
1337 |
+
Connection(248, 419),
|
1338 |
+
Connection(419, 197),
|
1339 |
+
Connection(197, 248),
|
1340 |
+
Connection(263, 255),
|
1341 |
+
Connection(255, 359),
|
1342 |
+
Connection(359, 263),
|
1343 |
+
Connection(440, 275),
|
1344 |
+
Connection(275, 274),
|
1345 |
+
Connection(274, 440),
|
1346 |
+
Connection(300, 383),
|
1347 |
+
Connection(383, 368),
|
1348 |
+
Connection(368, 300),
|
1349 |
+
Connection(351, 412),
|
1350 |
+
Connection(412, 465),
|
1351 |
+
Connection(465, 351),
|
1352 |
+
Connection(263, 467),
|
1353 |
+
Connection(467, 466),
|
1354 |
+
Connection(466, 263),
|
1355 |
+
Connection(301, 368),
|
1356 |
+
Connection(368, 389),
|
1357 |
+
Connection(389, 301),
|
1358 |
+
Connection(395, 378),
|
1359 |
+
Connection(378, 379),
|
1360 |
+
Connection(379, 395),
|
1361 |
+
Connection(412, 351),
|
1362 |
+
Connection(351, 419),
|
1363 |
+
Connection(419, 412),
|
1364 |
+
Connection(436, 426),
|
1365 |
+
Connection(426, 322),
|
1366 |
+
Connection(322, 436),
|
1367 |
+
Connection(2, 164),
|
1368 |
+
Connection(164, 393),
|
1369 |
+
Connection(393, 2),
|
1370 |
+
Connection(370, 462),
|
1371 |
+
Connection(462, 461),
|
1372 |
+
Connection(461, 370),
|
1373 |
+
Connection(164, 0),
|
1374 |
+
Connection(0, 267),
|
1375 |
+
Connection(267, 164),
|
1376 |
+
Connection(302, 11),
|
1377 |
+
Connection(11, 12),
|
1378 |
+
Connection(12, 302),
|
1379 |
+
Connection(268, 12),
|
1380 |
+
Connection(12, 13),
|
1381 |
+
Connection(13, 268),
|
1382 |
+
Connection(293, 300),
|
1383 |
+
Connection(300, 301),
|
1384 |
+
Connection(301, 293),
|
1385 |
+
Connection(446, 261),
|
1386 |
+
Connection(261, 340),
|
1387 |
+
Connection(340, 446),
|
1388 |
+
Connection(330, 266),
|
1389 |
+
Connection(266, 425),
|
1390 |
+
Connection(425, 330),
|
1391 |
+
Connection(426, 423),
|
1392 |
+
Connection(423, 391),
|
1393 |
+
Connection(391, 426),
|
1394 |
+
Connection(429, 355),
|
1395 |
+
Connection(355, 437),
|
1396 |
+
Connection(437, 429),
|
1397 |
+
Connection(391, 327),
|
1398 |
+
Connection(327, 326),
|
1399 |
+
Connection(326, 391),
|
1400 |
+
Connection(440, 457),
|
1401 |
+
Connection(457, 438),
|
1402 |
+
Connection(438, 440),
|
1403 |
+
Connection(341, 382),
|
1404 |
+
Connection(382, 362),
|
1405 |
+
Connection(362, 341),
|
1406 |
+
Connection(459, 457),
|
1407 |
+
Connection(457, 461),
|
1408 |
+
Connection(461, 459),
|
1409 |
+
Connection(434, 430),
|
1410 |
+
Connection(430, 394),
|
1411 |
+
Connection(394, 434),
|
1412 |
+
Connection(414, 463),
|
1413 |
+
Connection(463, 362),
|
1414 |
+
Connection(362, 414),
|
1415 |
+
Connection(396, 369),
|
1416 |
+
Connection(369, 262),
|
1417 |
+
Connection(262, 396),
|
1418 |
+
Connection(354, 461),
|
1419 |
+
Connection(461, 457),
|
1420 |
+
Connection(457, 354),
|
1421 |
+
Connection(316, 403),
|
1422 |
+
Connection(403, 402),
|
1423 |
+
Connection(402, 316),
|
1424 |
+
Connection(315, 404),
|
1425 |
+
Connection(404, 403),
|
1426 |
+
Connection(403, 315),
|
1427 |
+
Connection(314, 405),
|
1428 |
+
Connection(405, 404),
|
1429 |
+
Connection(404, 314),
|
1430 |
+
Connection(313, 406),
|
1431 |
+
Connection(406, 405),
|
1432 |
+
Connection(405, 313),
|
1433 |
+
Connection(421, 418),
|
1434 |
+
Connection(418, 406),
|
1435 |
+
Connection(406, 421),
|
1436 |
+
Connection(366, 401),
|
1437 |
+
Connection(401, 361),
|
1438 |
+
Connection(361, 366),
|
1439 |
+
Connection(306, 408),
|
1440 |
+
Connection(408, 407),
|
1441 |
+
Connection(407, 306),
|
1442 |
+
Connection(291, 409),
|
1443 |
+
Connection(409, 408),
|
1444 |
+
Connection(408, 291),
|
1445 |
+
Connection(287, 410),
|
1446 |
+
Connection(410, 409),
|
1447 |
+
Connection(409, 287),
|
1448 |
+
Connection(432, 436),
|
1449 |
+
Connection(436, 410),
|
1450 |
+
Connection(410, 432),
|
1451 |
+
Connection(434, 416),
|
1452 |
+
Connection(416, 411),
|
1453 |
+
Connection(411, 434),
|
1454 |
+
Connection(264, 368),
|
1455 |
+
Connection(368, 383),
|
1456 |
+
Connection(383, 264),
|
1457 |
+
Connection(309, 438),
|
1458 |
+
Connection(438, 457),
|
1459 |
+
Connection(457, 309),
|
1460 |
+
Connection(352, 376),
|
1461 |
+
Connection(376, 401),
|
1462 |
+
Connection(401, 352),
|
1463 |
+
Connection(274, 275),
|
1464 |
+
Connection(275, 4),
|
1465 |
+
Connection(4, 274),
|
1466 |
+
Connection(421, 428),
|
1467 |
+
Connection(428, 262),
|
1468 |
+
Connection(262, 421),
|
1469 |
+
Connection(294, 327),
|
1470 |
+
Connection(327, 358),
|
1471 |
+
Connection(358, 294),
|
1472 |
+
Connection(433, 416),
|
1473 |
+
Connection(416, 367),
|
1474 |
+
Connection(367, 433),
|
1475 |
+
Connection(289, 455),
|
1476 |
+
Connection(455, 439),
|
1477 |
+
Connection(439, 289),
|
1478 |
+
Connection(462, 370),
|
1479 |
+
Connection(370, 326),
|
1480 |
+
Connection(326, 462),
|
1481 |
+
Connection(2, 326),
|
1482 |
+
Connection(326, 370),
|
1483 |
+
Connection(370, 2),
|
1484 |
+
Connection(305, 460),
|
1485 |
+
Connection(460, 455),
|
1486 |
+
Connection(455, 305),
|
1487 |
+
Connection(254, 449),
|
1488 |
+
Connection(449, 448),
|
1489 |
+
Connection(448, 254),
|
1490 |
+
Connection(255, 261),
|
1491 |
+
Connection(261, 446),
|
1492 |
+
Connection(446, 255),
|
1493 |
+
Connection(253, 450),
|
1494 |
+
Connection(450, 449),
|
1495 |
+
Connection(449, 253),
|
1496 |
+
Connection(252, 451),
|
1497 |
+
Connection(451, 450),
|
1498 |
+
Connection(450, 252),
|
1499 |
+
Connection(256, 452),
|
1500 |
+
Connection(452, 451),
|
1501 |
+
Connection(451, 256),
|
1502 |
+
Connection(341, 453),
|
1503 |
+
Connection(453, 452),
|
1504 |
+
Connection(452, 341),
|
1505 |
+
Connection(413, 464),
|
1506 |
+
Connection(464, 463),
|
1507 |
+
Connection(463, 413),
|
1508 |
+
Connection(441, 413),
|
1509 |
+
Connection(413, 414),
|
1510 |
+
Connection(414, 441),
|
1511 |
+
Connection(258, 442),
|
1512 |
+
Connection(442, 441),
|
1513 |
+
Connection(441, 258),
|
1514 |
+
Connection(257, 443),
|
1515 |
+
Connection(443, 442),
|
1516 |
+
Connection(442, 257),
|
1517 |
+
Connection(259, 444),
|
1518 |
+
Connection(444, 443),
|
1519 |
+
Connection(443, 259),
|
1520 |
+
Connection(260, 445),
|
1521 |
+
Connection(445, 444),
|
1522 |
+
Connection(444, 260),
|
1523 |
+
Connection(467, 342),
|
1524 |
+
Connection(342, 445),
|
1525 |
+
Connection(445, 467),
|
1526 |
+
Connection(459, 458),
|
1527 |
+
Connection(458, 250),
|
1528 |
+
Connection(250, 459),
|
1529 |
+
Connection(289, 392),
|
1530 |
+
Connection(392, 290),
|
1531 |
+
Connection(290, 289),
|
1532 |
+
Connection(290, 328),
|
1533 |
+
Connection(328, 460),
|
1534 |
+
Connection(460, 290),
|
1535 |
+
Connection(376, 433),
|
1536 |
+
Connection(433, 435),
|
1537 |
+
Connection(435, 376),
|
1538 |
+
Connection(250, 290),
|
1539 |
+
Connection(290, 392),
|
1540 |
+
Connection(392, 250),
|
1541 |
+
Connection(411, 416),
|
1542 |
+
Connection(416, 433),
|
1543 |
+
Connection(433, 411),
|
1544 |
+
Connection(341, 463),
|
1545 |
+
Connection(463, 464),
|
1546 |
+
Connection(464, 341),
|
1547 |
+
Connection(453, 464),
|
1548 |
+
Connection(464, 465),
|
1549 |
+
Connection(465, 453),
|
1550 |
+
Connection(357, 465),
|
1551 |
+
Connection(465, 412),
|
1552 |
+
Connection(412, 357),
|
1553 |
+
Connection(343, 412),
|
1554 |
+
Connection(412, 399),
|
1555 |
+
Connection(399, 343),
|
1556 |
+
Connection(360, 363),
|
1557 |
+
Connection(363, 440),
|
1558 |
+
Connection(440, 360),
|
1559 |
+
Connection(437, 399),
|
1560 |
+
Connection(399, 456),
|
1561 |
+
Connection(456, 437),
|
1562 |
+
Connection(420, 456),
|
1563 |
+
Connection(456, 363),
|
1564 |
+
Connection(363, 420),
|
1565 |
+
Connection(401, 435),
|
1566 |
+
Connection(435, 288),
|
1567 |
+
Connection(288, 401),
|
1568 |
+
Connection(372, 383),
|
1569 |
+
Connection(383, 353),
|
1570 |
+
Connection(353, 372),
|
1571 |
+
Connection(339, 255),
|
1572 |
+
Connection(255, 249),
|
1573 |
+
Connection(249, 339),
|
1574 |
+
Connection(448, 261),
|
1575 |
+
Connection(261, 255),
|
1576 |
+
Connection(255, 448),
|
1577 |
+
Connection(133, 243),
|
1578 |
+
Connection(243, 190),
|
1579 |
+
Connection(190, 133),
|
1580 |
+
Connection(133, 155),
|
1581 |
+
Connection(155, 112),
|
1582 |
+
Connection(112, 133),
|
1583 |
+
Connection(33, 246),
|
1584 |
+
Connection(246, 247),
|
1585 |
+
Connection(247, 33),
|
1586 |
+
Connection(33, 130),
|
1587 |
+
Connection(130, 25),
|
1588 |
+
Connection(25, 33),
|
1589 |
+
Connection(398, 384),
|
1590 |
+
Connection(384, 286),
|
1591 |
+
Connection(286, 398),
|
1592 |
+
Connection(362, 398),
|
1593 |
+
Connection(398, 414),
|
1594 |
+
Connection(414, 362),
|
1595 |
+
Connection(362, 463),
|
1596 |
+
Connection(463, 341),
|
1597 |
+
Connection(341, 362),
|
1598 |
+
Connection(263, 359),
|
1599 |
+
Connection(359, 467),
|
1600 |
+
Connection(467, 263),
|
1601 |
+
Connection(263, 249),
|
1602 |
+
Connection(249, 255),
|
1603 |
+
Connection(255, 263),
|
1604 |
+
Connection(466, 467),
|
1605 |
+
Connection(467, 260),
|
1606 |
+
Connection(260, 466),
|
1607 |
+
Connection(75, 60),
|
1608 |
+
Connection(60, 166),
|
1609 |
+
Connection(166, 75),
|
1610 |
+
Connection(238, 239),
|
1611 |
+
Connection(239, 79),
|
1612 |
+
Connection(79, 238),
|
1613 |
+
Connection(162, 127),
|
1614 |
+
Connection(127, 139),
|
1615 |
+
Connection(139, 162),
|
1616 |
+
Connection(72, 11),
|
1617 |
+
Connection(11, 37),
|
1618 |
+
Connection(37, 72),
|
1619 |
+
Connection(121, 232),
|
1620 |
+
Connection(232, 120),
|
1621 |
+
Connection(120, 121),
|
1622 |
+
Connection(73, 72),
|
1623 |
+
Connection(72, 39),
|
1624 |
+
Connection(39, 73),
|
1625 |
+
Connection(114, 128),
|
1626 |
+
Connection(128, 47),
|
1627 |
+
Connection(47, 114),
|
1628 |
+
Connection(233, 232),
|
1629 |
+
Connection(232, 128),
|
1630 |
+
Connection(128, 233),
|
1631 |
+
Connection(103, 104),
|
1632 |
+
Connection(104, 67),
|
1633 |
+
Connection(67, 103),
|
1634 |
+
Connection(152, 175),
|
1635 |
+
Connection(175, 148),
|
1636 |
+
Connection(148, 152),
|
1637 |
+
Connection(119, 118),
|
1638 |
+
Connection(118, 101),
|
1639 |
+
Connection(101, 119),
|
1640 |
+
Connection(74, 73),
|
1641 |
+
Connection(73, 40),
|
1642 |
+
Connection(40, 74),
|
1643 |
+
Connection(107, 9),
|
1644 |
+
Connection(9, 108),
|
1645 |
+
Connection(108, 107),
|
1646 |
+
Connection(49, 48),
|
1647 |
+
Connection(48, 131),
|
1648 |
+
Connection(131, 49),
|
1649 |
+
Connection(32, 194),
|
1650 |
+
Connection(194, 211),
|
1651 |
+
Connection(211, 32),
|
1652 |
+
Connection(184, 74),
|
1653 |
+
Connection(74, 185),
|
1654 |
+
Connection(185, 184),
|
1655 |
+
Connection(191, 80),
|
1656 |
+
Connection(80, 183),
|
1657 |
+
Connection(183, 191),
|
1658 |
+
Connection(185, 40),
|
1659 |
+
Connection(40, 186),
|
1660 |
+
Connection(186, 185),
|
1661 |
+
Connection(119, 230),
|
1662 |
+
Connection(230, 118),
|
1663 |
+
Connection(118, 119),
|
1664 |
+
Connection(210, 202),
|
1665 |
+
Connection(202, 214),
|
1666 |
+
Connection(214, 210),
|
1667 |
+
Connection(84, 83),
|
1668 |
+
Connection(83, 17),
|
1669 |
+
Connection(17, 84),
|
1670 |
+
Connection(77, 76),
|
1671 |
+
Connection(76, 146),
|
1672 |
+
Connection(146, 77),
|
1673 |
+
Connection(161, 160),
|
1674 |
+
Connection(160, 30),
|
1675 |
+
Connection(30, 161),
|
1676 |
+
Connection(190, 56),
|
1677 |
+
Connection(56, 173),
|
1678 |
+
Connection(173, 190),
|
1679 |
+
Connection(182, 106),
|
1680 |
+
Connection(106, 194),
|
1681 |
+
Connection(194, 182),
|
1682 |
+
Connection(138, 135),
|
1683 |
+
Connection(135, 192),
|
1684 |
+
Connection(192, 138),
|
1685 |
+
Connection(129, 203),
|
1686 |
+
Connection(203, 98),
|
1687 |
+
Connection(98, 129),
|
1688 |
+
Connection(54, 21),
|
1689 |
+
Connection(21, 68),
|
1690 |
+
Connection(68, 54),
|
1691 |
+
Connection(5, 51),
|
1692 |
+
Connection(51, 4),
|
1693 |
+
Connection(4, 5),
|
1694 |
+
Connection(145, 144),
|
1695 |
+
Connection(144, 23),
|
1696 |
+
Connection(23, 145),
|
1697 |
+
Connection(90, 77),
|
1698 |
+
Connection(77, 91),
|
1699 |
+
Connection(91, 90),
|
1700 |
+
Connection(207, 205),
|
1701 |
+
Connection(205, 187),
|
1702 |
+
Connection(187, 207),
|
1703 |
+
Connection(83, 201),
|
1704 |
+
Connection(201, 18),
|
1705 |
+
Connection(18, 83),
|
1706 |
+
Connection(181, 91),
|
1707 |
+
Connection(91, 182),
|
1708 |
+
Connection(182, 181),
|
1709 |
+
Connection(180, 90),
|
1710 |
+
Connection(90, 181),
|
1711 |
+
Connection(181, 180),
|
1712 |
+
Connection(16, 85),
|
1713 |
+
Connection(85, 17),
|
1714 |
+
Connection(17, 16),
|
1715 |
+
Connection(205, 206),
|
1716 |
+
Connection(206, 36),
|
1717 |
+
Connection(36, 205),
|
1718 |
+
Connection(176, 148),
|
1719 |
+
Connection(148, 140),
|
1720 |
+
Connection(140, 176),
|
1721 |
+
Connection(165, 92),
|
1722 |
+
Connection(92, 39),
|
1723 |
+
Connection(39, 165),
|
1724 |
+
Connection(245, 193),
|
1725 |
+
Connection(193, 244),
|
1726 |
+
Connection(244, 245),
|
1727 |
+
Connection(27, 159),
|
1728 |
+
Connection(159, 28),
|
1729 |
+
Connection(28, 27),
|
1730 |
+
Connection(30, 247),
|
1731 |
+
Connection(247, 161),
|
1732 |
+
Connection(161, 30),
|
1733 |
+
Connection(174, 236),
|
1734 |
+
Connection(236, 196),
|
1735 |
+
Connection(196, 174),
|
1736 |
+
Connection(103, 54),
|
1737 |
+
Connection(54, 104),
|
1738 |
+
Connection(104, 103),
|
1739 |
+
Connection(55, 193),
|
1740 |
+
Connection(193, 8),
|
1741 |
+
Connection(8, 55),
|
1742 |
+
Connection(111, 117),
|
1743 |
+
Connection(117, 31),
|
1744 |
+
Connection(31, 111),
|
1745 |
+
Connection(221, 189),
|
1746 |
+
Connection(189, 55),
|
1747 |
+
Connection(55, 221),
|
1748 |
+
Connection(240, 98),
|
1749 |
+
Connection(98, 99),
|
1750 |
+
Connection(99, 240),
|
1751 |
+
Connection(142, 126),
|
1752 |
+
Connection(126, 100),
|
1753 |
+
Connection(100, 142),
|
1754 |
+
Connection(219, 166),
|
1755 |
+
Connection(166, 218),
|
1756 |
+
Connection(218, 219),
|
1757 |
+
Connection(112, 155),
|
1758 |
+
Connection(155, 26),
|
1759 |
+
Connection(26, 112),
|
1760 |
+
Connection(198, 209),
|
1761 |
+
Connection(209, 131),
|
1762 |
+
Connection(131, 198),
|
1763 |
+
Connection(169, 135),
|
1764 |
+
Connection(135, 150),
|
1765 |
+
Connection(150, 169),
|
1766 |
+
Connection(114, 47),
|
1767 |
+
Connection(47, 217),
|
1768 |
+
Connection(217, 114),
|
1769 |
+
Connection(224, 223),
|
1770 |
+
Connection(223, 53),
|
1771 |
+
Connection(53, 224),
|
1772 |
+
Connection(220, 45),
|
1773 |
+
Connection(45, 134),
|
1774 |
+
Connection(134, 220),
|
1775 |
+
Connection(32, 211),
|
1776 |
+
Connection(211, 140),
|
1777 |
+
Connection(140, 32),
|
1778 |
+
Connection(109, 67),
|
1779 |
+
Connection(67, 108),
|
1780 |
+
Connection(108, 109),
|
1781 |
+
Connection(146, 43),
|
1782 |
+
Connection(43, 91),
|
1783 |
+
Connection(91, 146),
|
1784 |
+
Connection(231, 230),
|
1785 |
+
Connection(230, 120),
|
1786 |
+
Connection(120, 231),
|
1787 |
+
Connection(113, 226),
|
1788 |
+
Connection(226, 247),
|
1789 |
+
Connection(247, 113),
|
1790 |
+
Connection(105, 63),
|
1791 |
+
Connection(63, 52),
|
1792 |
+
Connection(52, 105),
|
1793 |
+
Connection(241, 238),
|
1794 |
+
Connection(238, 242),
|
1795 |
+
Connection(242, 241),
|
1796 |
+
Connection(124, 46),
|
1797 |
+
Connection(46, 156),
|
1798 |
+
Connection(156, 124),
|
1799 |
+
Connection(95, 78),
|
1800 |
+
Connection(78, 96),
|
1801 |
+
Connection(96, 95),
|
1802 |
+
Connection(70, 46),
|
1803 |
+
Connection(46, 63),
|
1804 |
+
Connection(63, 70),
|
1805 |
+
Connection(116, 143),
|
1806 |
+
Connection(143, 227),
|
1807 |
+
Connection(227, 116),
|
1808 |
+
Connection(116, 123),
|
1809 |
+
Connection(123, 111),
|
1810 |
+
Connection(111, 116),
|
1811 |
+
Connection(1, 44),
|
1812 |
+
Connection(44, 19),
|
1813 |
+
Connection(19, 1),
|
1814 |
+
Connection(3, 236),
|
1815 |
+
Connection(236, 51),
|
1816 |
+
Connection(51, 3),
|
1817 |
+
Connection(207, 216),
|
1818 |
+
Connection(216, 205),
|
1819 |
+
Connection(205, 207),
|
1820 |
+
Connection(26, 154),
|
1821 |
+
Connection(154, 22),
|
1822 |
+
Connection(22, 26),
|
1823 |
+
Connection(165, 39),
|
1824 |
+
Connection(39, 167),
|
1825 |
+
Connection(167, 165),
|
1826 |
+
Connection(199, 200),
|
1827 |
+
Connection(200, 208),
|
1828 |
+
Connection(208, 199),
|
1829 |
+
Connection(101, 36),
|
1830 |
+
Connection(36, 100),
|
1831 |
+
Connection(100, 101),
|
1832 |
+
Connection(43, 57),
|
1833 |
+
Connection(57, 202),
|
1834 |
+
Connection(202, 43),
|
1835 |
+
Connection(242, 20),
|
1836 |
+
Connection(20, 99),
|
1837 |
+
Connection(99, 242),
|
1838 |
+
Connection(56, 28),
|
1839 |
+
Connection(28, 157),
|
1840 |
+
Connection(157, 56),
|
1841 |
+
Connection(124, 35),
|
1842 |
+
Connection(35, 113),
|
1843 |
+
Connection(113, 124),
|
1844 |
+
Connection(29, 160),
|
1845 |
+
Connection(160, 27),
|
1846 |
+
Connection(27, 29),
|
1847 |
+
Connection(211, 204),
|
1848 |
+
Connection(204, 210),
|
1849 |
+
Connection(210, 211),
|
1850 |
+
Connection(124, 113),
|
1851 |
+
Connection(113, 46),
|
1852 |
+
Connection(46, 124),
|
1853 |
+
Connection(106, 43),
|
1854 |
+
Connection(43, 204),
|
1855 |
+
Connection(204, 106),
|
1856 |
+
Connection(96, 62),
|
1857 |
+
Connection(62, 77),
|
1858 |
+
Connection(77, 96),
|
1859 |
+
Connection(227, 137),
|
1860 |
+
Connection(137, 116),
|
1861 |
+
Connection(116, 227),
|
1862 |
+
Connection(73, 41),
|
1863 |
+
Connection(41, 72),
|
1864 |
+
Connection(72, 73),
|
1865 |
+
Connection(36, 203),
|
1866 |
+
Connection(203, 142),
|
1867 |
+
Connection(142, 36),
|
1868 |
+
Connection(235, 64),
|
1869 |
+
Connection(64, 240),
|
1870 |
+
Connection(240, 235),
|
1871 |
+
Connection(48, 49),
|
1872 |
+
Connection(49, 64),
|
1873 |
+
Connection(64, 48),
|
1874 |
+
Connection(42, 41),
|
1875 |
+
Connection(41, 74),
|
1876 |
+
Connection(74, 42),
|
1877 |
+
Connection(214, 212),
|
1878 |
+
Connection(212, 207),
|
1879 |
+
Connection(207, 214),
|
1880 |
+
Connection(183, 42),
|
1881 |
+
Connection(42, 184),
|
1882 |
+
Connection(184, 183),
|
1883 |
+
Connection(210, 169),
|
1884 |
+
Connection(169, 211),
|
1885 |
+
Connection(211, 210),
|
1886 |
+
Connection(140, 170),
|
1887 |
+
Connection(170, 176),
|
1888 |
+
Connection(176, 140),
|
1889 |
+
Connection(104, 105),
|
1890 |
+
Connection(105, 69),
|
1891 |
+
Connection(69, 104),
|
1892 |
+
Connection(193, 122),
|
1893 |
+
Connection(122, 168),
|
1894 |
+
Connection(168, 193),
|
1895 |
+
Connection(50, 123),
|
1896 |
+
Connection(123, 187),
|
1897 |
+
Connection(187, 50),
|
1898 |
+
Connection(89, 96),
|
1899 |
+
Connection(96, 90),
|
1900 |
+
Connection(90, 89),
|
1901 |
+
Connection(66, 65),
|
1902 |
+
Connection(65, 107),
|
1903 |
+
Connection(107, 66),
|
1904 |
+
Connection(179, 89),
|
1905 |
+
Connection(89, 180),
|
1906 |
+
Connection(180, 179),
|
1907 |
+
Connection(119, 101),
|
1908 |
+
Connection(101, 120),
|
1909 |
+
Connection(120, 119),
|
1910 |
+
Connection(68, 63),
|
1911 |
+
Connection(63, 104),
|
1912 |
+
Connection(104, 68),
|
1913 |
+
Connection(234, 93),
|
1914 |
+
Connection(93, 227),
|
1915 |
+
Connection(227, 234),
|
1916 |
+
Connection(16, 15),
|
1917 |
+
Connection(15, 85),
|
1918 |
+
Connection(85, 16),
|
1919 |
+
Connection(209, 129),
|
1920 |
+
Connection(129, 49),
|
1921 |
+
Connection(49, 209),
|
1922 |
+
Connection(15, 14),
|
1923 |
+
Connection(14, 86),
|
1924 |
+
Connection(86, 15),
|
1925 |
+
Connection(107, 55),
|
1926 |
+
Connection(55, 9),
|
1927 |
+
Connection(9, 107),
|
1928 |
+
Connection(120, 100),
|
1929 |
+
Connection(100, 121),
|
1930 |
+
Connection(121, 120),
|
1931 |
+
Connection(153, 145),
|
1932 |
+
Connection(145, 22),
|
1933 |
+
Connection(22, 153),
|
1934 |
+
Connection(178, 88),
|
1935 |
+
Connection(88, 179),
|
1936 |
+
Connection(179, 178),
|
1937 |
+
Connection(197, 6),
|
1938 |
+
Connection(6, 196),
|
1939 |
+
Connection(196, 197),
|
1940 |
+
Connection(89, 88),
|
1941 |
+
Connection(88, 96),
|
1942 |
+
Connection(96, 89),
|
1943 |
+
Connection(135, 138),
|
1944 |
+
Connection(138, 136),
|
1945 |
+
Connection(136, 135),
|
1946 |
+
Connection(138, 215),
|
1947 |
+
Connection(215, 172),
|
1948 |
+
Connection(172, 138),
|
1949 |
+
Connection(218, 115),
|
1950 |
+
Connection(115, 219),
|
1951 |
+
Connection(219, 218),
|
1952 |
+
Connection(41, 42),
|
1953 |
+
Connection(42, 81),
|
1954 |
+
Connection(81, 41),
|
1955 |
+
Connection(5, 195),
|
1956 |
+
Connection(195, 51),
|
1957 |
+
Connection(51, 5),
|
1958 |
+
Connection(57, 43),
|
1959 |
+
Connection(43, 61),
|
1960 |
+
Connection(61, 57),
|
1961 |
+
Connection(208, 171),
|
1962 |
+
Connection(171, 199),
|
1963 |
+
Connection(199, 208),
|
1964 |
+
Connection(41, 81),
|
1965 |
+
Connection(81, 38),
|
1966 |
+
Connection(38, 41),
|
1967 |
+
Connection(224, 53),
|
1968 |
+
Connection(53, 225),
|
1969 |
+
Connection(225, 224),
|
1970 |
+
Connection(24, 144),
|
1971 |
+
Connection(144, 110),
|
1972 |
+
Connection(110, 24),
|
1973 |
+
Connection(105, 52),
|
1974 |
+
Connection(52, 66),
|
1975 |
+
Connection(66, 105),
|
1976 |
+
Connection(118, 229),
|
1977 |
+
Connection(229, 117),
|
1978 |
+
Connection(117, 118),
|
1979 |
+
Connection(227, 34),
|
1980 |
+
Connection(34, 234),
|
1981 |
+
Connection(234, 227),
|
1982 |
+
Connection(66, 107),
|
1983 |
+
Connection(107, 69),
|
1984 |
+
Connection(69, 66),
|
1985 |
+
Connection(10, 109),
|
1986 |
+
Connection(109, 151),
|
1987 |
+
Connection(151, 10),
|
1988 |
+
Connection(219, 48),
|
1989 |
+
Connection(48, 235),
|
1990 |
+
Connection(235, 219),
|
1991 |
+
Connection(183, 62),
|
1992 |
+
Connection(62, 191),
|
1993 |
+
Connection(191, 183),
|
1994 |
+
Connection(142, 129),
|
1995 |
+
Connection(129, 126),
|
1996 |
+
Connection(126, 142),
|
1997 |
+
Connection(116, 111),
|
1998 |
+
Connection(111, 143),
|
1999 |
+
Connection(143, 116),
|
2000 |
+
Connection(118, 117),
|
2001 |
+
Connection(117, 50),
|
2002 |
+
Connection(50, 118),
|
2003 |
+
Connection(223, 222),
|
2004 |
+
Connection(222, 52),
|
2005 |
+
Connection(52, 223),
|
2006 |
+
Connection(94, 19),
|
2007 |
+
Connection(19, 141),
|
2008 |
+
Connection(141, 94),
|
2009 |
+
Connection(222, 221),
|
2010 |
+
Connection(221, 65),
|
2011 |
+
Connection(65, 222),
|
2012 |
+
Connection(196, 3),
|
2013 |
+
Connection(3, 197),
|
2014 |
+
Connection(197, 196),
|
2015 |
+
Connection(45, 220),
|
2016 |
+
Connection(220, 44),
|
2017 |
+
Connection(44, 45),
|
2018 |
+
Connection(156, 70),
|
2019 |
+
Connection(70, 139),
|
2020 |
+
Connection(139, 156),
|
2021 |
+
Connection(188, 122),
|
2022 |
+
Connection(122, 245),
|
2023 |
+
Connection(245, 188),
|
2024 |
+
Connection(139, 71),
|
2025 |
+
Connection(71, 162),
|
2026 |
+
Connection(162, 139),
|
2027 |
+
Connection(149, 170),
|
2028 |
+
Connection(170, 150),
|
2029 |
+
Connection(150, 149),
|
2030 |
+
Connection(122, 188),
|
2031 |
+
Connection(188, 196),
|
2032 |
+
Connection(196, 122),
|
2033 |
+
Connection(206, 216),
|
2034 |
+
Connection(216, 92),
|
2035 |
+
Connection(92, 206),
|
2036 |
+
Connection(164, 2),
|
2037 |
+
Connection(2, 167),
|
2038 |
+
Connection(167, 164),
|
2039 |
+
Connection(242, 141),
|
2040 |
+
Connection(141, 241),
|
2041 |
+
Connection(241, 242),
|
2042 |
+
Connection(0, 164),
|
2043 |
+
Connection(164, 37),
|
2044 |
+
Connection(37, 0),
|
2045 |
+
Connection(11, 72),
|
2046 |
+
Connection(72, 12),
|
2047 |
+
Connection(12, 11),
|
2048 |
+
Connection(12, 38),
|
2049 |
+
Connection(38, 13),
|
2050 |
+
Connection(13, 12),
|
2051 |
+
Connection(70, 63),
|
2052 |
+
Connection(63, 71),
|
2053 |
+
Connection(71, 70),
|
2054 |
+
Connection(31, 226),
|
2055 |
+
Connection(226, 111),
|
2056 |
+
Connection(111, 31),
|
2057 |
+
Connection(36, 101),
|
2058 |
+
Connection(101, 205),
|
2059 |
+
Connection(205, 36),
|
2060 |
+
Connection(203, 206),
|
2061 |
+
Connection(206, 165),
|
2062 |
+
Connection(165, 203),
|
2063 |
+
Connection(126, 209),
|
2064 |
+
Connection(209, 217),
|
2065 |
+
Connection(217, 126),
|
2066 |
+
Connection(98, 165),
|
2067 |
+
Connection(165, 97),
|
2068 |
+
Connection(97, 98),
|
2069 |
+
Connection(237, 220),
|
2070 |
+
Connection(220, 218),
|
2071 |
+
Connection(218, 237),
|
2072 |
+
Connection(237, 239),
|
2073 |
+
Connection(239, 241),
|
2074 |
+
Connection(241, 237),
|
2075 |
+
Connection(210, 214),
|
2076 |
+
Connection(214, 169),
|
2077 |
+
Connection(169, 210),
|
2078 |
+
Connection(140, 171),
|
2079 |
+
Connection(171, 32),
|
2080 |
+
Connection(32, 140),
|
2081 |
+
Connection(241, 125),
|
2082 |
+
Connection(125, 237),
|
2083 |
+
Connection(237, 241),
|
2084 |
+
Connection(179, 86),
|
2085 |
+
Connection(86, 178),
|
2086 |
+
Connection(178, 179),
|
2087 |
+
Connection(180, 85),
|
2088 |
+
Connection(85, 179),
|
2089 |
+
Connection(179, 180),
|
2090 |
+
Connection(181, 84),
|
2091 |
+
Connection(84, 180),
|
2092 |
+
Connection(180, 181),
|
2093 |
+
Connection(182, 83),
|
2094 |
+
Connection(83, 181),
|
2095 |
+
Connection(181, 182),
|
2096 |
+
Connection(194, 201),
|
2097 |
+
Connection(201, 182),
|
2098 |
+
Connection(182, 194),
|
2099 |
+
Connection(177, 137),
|
2100 |
+
Connection(137, 132),
|
2101 |
+
Connection(132, 177),
|
2102 |
+
Connection(184, 76),
|
2103 |
+
Connection(76, 183),
|
2104 |
+
Connection(183, 184),
|
2105 |
+
Connection(185, 61),
|
2106 |
+
Connection(61, 184),
|
2107 |
+
Connection(184, 185),
|
2108 |
+
Connection(186, 57),
|
2109 |
+
Connection(57, 185),
|
2110 |
+
Connection(185, 186),
|
2111 |
+
Connection(216, 212),
|
2112 |
+
Connection(212, 186),
|
2113 |
+
Connection(186, 216),
|
2114 |
+
Connection(192, 214),
|
2115 |
+
Connection(214, 187),
|
2116 |
+
Connection(187, 192),
|
2117 |
+
Connection(139, 34),
|
2118 |
+
Connection(34, 156),
|
2119 |
+
Connection(156, 139),
|
2120 |
+
Connection(218, 79),
|
2121 |
+
Connection(79, 237),
|
2122 |
+
Connection(237, 218),
|
2123 |
+
Connection(147, 123),
|
2124 |
+
Connection(123, 177),
|
2125 |
+
Connection(177, 147),
|
2126 |
+
Connection(45, 44),
|
2127 |
+
Connection(44, 4),
|
2128 |
+
Connection(4, 45),
|
2129 |
+
Connection(208, 201),
|
2130 |
+
Connection(201, 32),
|
2131 |
+
Connection(32, 208),
|
2132 |
+
Connection(98, 64),
|
2133 |
+
Connection(64, 129),
|
2134 |
+
Connection(129, 98),
|
2135 |
+
Connection(192, 213),
|
2136 |
+
Connection(213, 138),
|
2137 |
+
Connection(138, 192),
|
2138 |
+
Connection(235, 59),
|
2139 |
+
Connection(59, 219),
|
2140 |
+
Connection(219, 235),
|
2141 |
+
Connection(141, 242),
|
2142 |
+
Connection(242, 97),
|
2143 |
+
Connection(97, 141),
|
2144 |
+
Connection(97, 2),
|
2145 |
+
Connection(2, 141),
|
2146 |
+
Connection(141, 97),
|
2147 |
+
Connection(240, 75),
|
2148 |
+
Connection(75, 235),
|
2149 |
+
Connection(235, 240),
|
2150 |
+
Connection(229, 24),
|
2151 |
+
Connection(24, 228),
|
2152 |
+
Connection(228, 229),
|
2153 |
+
Connection(31, 25),
|
2154 |
+
Connection(25, 226),
|
2155 |
+
Connection(226, 31),
|
2156 |
+
Connection(230, 23),
|
2157 |
+
Connection(23, 229),
|
2158 |
+
Connection(229, 230),
|
2159 |
+
Connection(231, 22),
|
2160 |
+
Connection(22, 230),
|
2161 |
+
Connection(230, 231),
|
2162 |
+
Connection(232, 26),
|
2163 |
+
Connection(26, 231),
|
2164 |
+
Connection(231, 232),
|
2165 |
+
Connection(233, 112),
|
2166 |
+
Connection(112, 232),
|
2167 |
+
Connection(232, 233),
|
2168 |
+
Connection(244, 189),
|
2169 |
+
Connection(189, 243),
|
2170 |
+
Connection(243, 244),
|
2171 |
+
Connection(189, 221),
|
2172 |
+
Connection(221, 190),
|
2173 |
+
Connection(190, 189),
|
2174 |
+
Connection(222, 28),
|
2175 |
+
Connection(28, 221),
|
2176 |
+
Connection(221, 222),
|
2177 |
+
Connection(223, 27),
|
2178 |
+
Connection(27, 222),
|
2179 |
+
Connection(222, 223),
|
2180 |
+
Connection(224, 29),
|
2181 |
+
Connection(29, 223),
|
2182 |
+
Connection(223, 224),
|
2183 |
+
Connection(225, 30),
|
2184 |
+
Connection(30, 224),
|
2185 |
+
Connection(224, 225),
|
2186 |
+
Connection(113, 247),
|
2187 |
+
Connection(247, 225),
|
2188 |
+
Connection(225, 113),
|
2189 |
+
Connection(99, 60),
|
2190 |
+
Connection(60, 240),
|
2191 |
+
Connection(240, 99),
|
2192 |
+
Connection(213, 147),
|
2193 |
+
Connection(147, 215),
|
2194 |
+
Connection(215, 213),
|
2195 |
+
Connection(60, 20),
|
2196 |
+
Connection(20, 166),
|
2197 |
+
Connection(166, 60),
|
2198 |
+
Connection(192, 187),
|
2199 |
+
Connection(187, 213),
|
2200 |
+
Connection(213, 192),
|
2201 |
+
Connection(243, 112),
|
2202 |
+
Connection(112, 244),
|
2203 |
+
Connection(244, 243),
|
2204 |
+
Connection(244, 233),
|
2205 |
+
Connection(233, 245),
|
2206 |
+
Connection(245, 244),
|
2207 |
+
Connection(245, 128),
|
2208 |
+
Connection(128, 188),
|
2209 |
+
Connection(188, 245),
|
2210 |
+
Connection(188, 114),
|
2211 |
+
Connection(114, 174),
|
2212 |
+
Connection(174, 188),
|
2213 |
+
Connection(134, 131),
|
2214 |
+
Connection(131, 220),
|
2215 |
+
Connection(220, 134),
|
2216 |
+
Connection(174, 217),
|
2217 |
+
Connection(217, 236),
|
2218 |
+
Connection(236, 174),
|
2219 |
+
Connection(236, 198),
|
2220 |
+
Connection(198, 134),
|
2221 |
+
Connection(134, 236),
|
2222 |
+
Connection(215, 177),
|
2223 |
+
Connection(177, 58),
|
2224 |
+
Connection(58, 215),
|
2225 |
+
Connection(156, 143),
|
2226 |
+
Connection(143, 124),
|
2227 |
+
Connection(124, 156),
|
2228 |
+
Connection(25, 110),
|
2229 |
+
Connection(110, 7),
|
2230 |
+
Connection(7, 25),
|
2231 |
+
Connection(31, 228),
|
2232 |
+
Connection(228, 25),
|
2233 |
+
Connection(25, 31),
|
2234 |
+
Connection(264, 356),
|
2235 |
+
Connection(356, 368),
|
2236 |
+
Connection(368, 264),
|
2237 |
+
Connection(0, 11),
|
2238 |
+
Connection(11, 267),
|
2239 |
+
Connection(267, 0),
|
2240 |
+
Connection(451, 452),
|
2241 |
+
Connection(452, 349),
|
2242 |
+
Connection(349, 451),
|
2243 |
+
Connection(267, 302),
|
2244 |
+
Connection(302, 269),
|
2245 |
+
Connection(269, 267),
|
2246 |
+
Connection(350, 357),
|
2247 |
+
Connection(357, 277),
|
2248 |
+
Connection(277, 350),
|
2249 |
+
Connection(350, 452),
|
2250 |
+
Connection(452, 357),
|
2251 |
+
Connection(357, 350),
|
2252 |
+
Connection(299, 333),
|
2253 |
+
Connection(333, 297),
|
2254 |
+
Connection(297, 299),
|
2255 |
+
Connection(396, 175),
|
2256 |
+
Connection(175, 377),
|
2257 |
+
Connection(377, 396),
|
2258 |
+
Connection(280, 347),
|
2259 |
+
Connection(347, 330),
|
2260 |
+
Connection(330, 280),
|
2261 |
+
Connection(269, 303),
|
2262 |
+
Connection(303, 270),
|
2263 |
+
Connection(270, 269),
|
2264 |
+
Connection(151, 9),
|
2265 |
+
Connection(9, 337),
|
2266 |
+
Connection(337, 151),
|
2267 |
+
Connection(344, 278),
|
2268 |
+
Connection(278, 360),
|
2269 |
+
Connection(360, 344),
|
2270 |
+
Connection(424, 418),
|
2271 |
+
Connection(418, 431),
|
2272 |
+
Connection(431, 424),
|
2273 |
+
Connection(270, 304),
|
2274 |
+
Connection(304, 409),
|
2275 |
+
Connection(409, 270),
|
2276 |
+
Connection(272, 310),
|
2277 |
+
Connection(310, 407),
|
2278 |
+
Connection(407, 272),
|
2279 |
+
Connection(322, 270),
|
2280 |
+
Connection(270, 410),
|
2281 |
+
Connection(410, 322),
|
2282 |
+
Connection(449, 450),
|
2283 |
+
Connection(450, 347),
|
2284 |
+
Connection(347, 449),
|
2285 |
+
Connection(432, 422),
|
2286 |
+
Connection(422, 434),
|
2287 |
+
Connection(434, 432),
|
2288 |
+
Connection(18, 313),
|
2289 |
+
Connection(313, 17),
|
2290 |
+
Connection(17, 18),
|
2291 |
+
Connection(291, 306),
|
2292 |
+
Connection(306, 375),
|
2293 |
+
Connection(375, 291),
|
2294 |
+
Connection(259, 387),
|
2295 |
+
Connection(387, 260),
|
2296 |
+
Connection(260, 259),
|
2297 |
+
Connection(424, 335),
|
2298 |
+
Connection(335, 418),
|
2299 |
+
Connection(418, 424),
|
2300 |
+
Connection(434, 364),
|
2301 |
+
Connection(364, 416),
|
2302 |
+
Connection(416, 434),
|
2303 |
+
Connection(391, 423),
|
2304 |
+
Connection(423, 327),
|
2305 |
+
Connection(327, 391),
|
2306 |
+
Connection(301, 251),
|
2307 |
+
Connection(251, 298),
|
2308 |
+
Connection(298, 301),
|
2309 |
+
Connection(275, 281),
|
2310 |
+
Connection(281, 4),
|
2311 |
+
Connection(4, 275),
|
2312 |
+
Connection(254, 373),
|
2313 |
+
Connection(373, 253),
|
2314 |
+
Connection(253, 254),
|
2315 |
+
Connection(375, 307),
|
2316 |
+
Connection(307, 321),
|
2317 |
+
Connection(321, 375),
|
2318 |
+
Connection(280, 425),
|
2319 |
+
Connection(425, 411),
|
2320 |
+
Connection(411, 280),
|
2321 |
+
Connection(200, 421),
|
2322 |
+
Connection(421, 18),
|
2323 |
+
Connection(18, 200),
|
2324 |
+
Connection(335, 321),
|
2325 |
+
Connection(321, 406),
|
2326 |
+
Connection(406, 335),
|
2327 |
+
Connection(321, 320),
|
2328 |
+
Connection(320, 405),
|
2329 |
+
Connection(405, 321),
|
2330 |
+
Connection(314, 315),
|
2331 |
+
Connection(315, 17),
|
2332 |
+
Connection(17, 314),
|
2333 |
+
Connection(423, 426),
|
2334 |
+
Connection(426, 266),
|
2335 |
+
Connection(266, 423),
|
2336 |
+
Connection(396, 377),
|
2337 |
+
Connection(377, 369),
|
2338 |
+
Connection(369, 396),
|
2339 |
+
Connection(270, 322),
|
2340 |
+
Connection(322, 269),
|
2341 |
+
Connection(269, 270),
|
2342 |
+
Connection(413, 417),
|
2343 |
+
Connection(417, 464),
|
2344 |
+
Connection(464, 413),
|
2345 |
+
Connection(385, 386),
|
2346 |
+
Connection(386, 258),
|
2347 |
+
Connection(258, 385),
|
2348 |
+
Connection(248, 456),
|
2349 |
+
Connection(456, 419),
|
2350 |
+
Connection(419, 248),
|
2351 |
+
Connection(298, 284),
|
2352 |
+
Connection(284, 333),
|
2353 |
+
Connection(333, 298),
|
2354 |
+
Connection(168, 417),
|
2355 |
+
Connection(417, 8),
|
2356 |
+
Connection(8, 168),
|
2357 |
+
Connection(448, 346),
|
2358 |
+
Connection(346, 261),
|
2359 |
+
Connection(261, 448),
|
2360 |
+
Connection(417, 413),
|
2361 |
+
Connection(413, 285),
|
2362 |
+
Connection(285, 417),
|
2363 |
+
Connection(326, 327),
|
2364 |
+
Connection(327, 328),
|
2365 |
+
Connection(328, 326),
|
2366 |
+
Connection(277, 355),
|
2367 |
+
Connection(355, 329),
|
2368 |
+
Connection(329, 277),
|
2369 |
+
Connection(309, 392),
|
2370 |
+
Connection(392, 438),
|
2371 |
+
Connection(438, 309),
|
2372 |
+
Connection(381, 382),
|
2373 |
+
Connection(382, 256),
|
2374 |
+
Connection(256, 381),
|
2375 |
+
Connection(279, 429),
|
2376 |
+
Connection(429, 360),
|
2377 |
+
Connection(360, 279),
|
2378 |
+
Connection(365, 364),
|
2379 |
+
Connection(364, 379),
|
2380 |
+
Connection(379, 365),
|
2381 |
+
Connection(355, 277),
|
2382 |
+
Connection(277, 437),
|
2383 |
+
Connection(437, 355),
|
2384 |
+
Connection(282, 443),
|
2385 |
+
Connection(443, 283),
|
2386 |
+
Connection(283, 282),
|
2387 |
+
Connection(281, 275),
|
2388 |
+
Connection(275, 363),
|
2389 |
+
Connection(363, 281),
|
2390 |
+
Connection(395, 431),
|
2391 |
+
Connection(431, 369),
|
2392 |
+
Connection(369, 395),
|
2393 |
+
Connection(299, 297),
|
2394 |
+
Connection(297, 337),
|
2395 |
+
Connection(337, 299),
|
2396 |
+
Connection(335, 273),
|
2397 |
+
Connection(273, 321),
|
2398 |
+
Connection(321, 335),
|
2399 |
+
Connection(348, 450),
|
2400 |
+
Connection(450, 349),
|
2401 |
+
Connection(349, 348),
|
2402 |
+
Connection(359, 446),
|
2403 |
+
Connection(446, 467),
|
2404 |
+
Connection(467, 359),
|
2405 |
+
Connection(283, 293),
|
2406 |
+
Connection(293, 282),
|
2407 |
+
Connection(282, 283),
|
2408 |
+
Connection(250, 458),
|
2409 |
+
Connection(458, 462),
|
2410 |
+
Connection(462, 250),
|
2411 |
+
Connection(300, 276),
|
2412 |
+
Connection(276, 383),
|
2413 |
+
Connection(383, 300),
|
2414 |
+
Connection(292, 308),
|
2415 |
+
Connection(308, 325),
|
2416 |
+
Connection(325, 292),
|
2417 |
+
Connection(283, 276),
|
2418 |
+
Connection(276, 293),
|
2419 |
+
Connection(293, 283),
|
2420 |
+
Connection(264, 372),
|
2421 |
+
Connection(372, 447),
|
2422 |
+
Connection(447, 264),
|
2423 |
+
Connection(346, 352),
|
2424 |
+
Connection(352, 340),
|
2425 |
+
Connection(340, 346),
|
2426 |
+
Connection(354, 274),
|
2427 |
+
Connection(274, 19),
|
2428 |
+
Connection(19, 354),
|
2429 |
+
Connection(363, 456),
|
2430 |
+
Connection(456, 281),
|
2431 |
+
Connection(281, 363),
|
2432 |
+
Connection(426, 436),
|
2433 |
+
Connection(436, 425),
|
2434 |
+
Connection(425, 426),
|
2435 |
+
Connection(380, 381),
|
2436 |
+
Connection(381, 252),
|
2437 |
+
Connection(252, 380),
|
2438 |
+
Connection(267, 269),
|
2439 |
+
Connection(269, 393),
|
2440 |
+
Connection(393, 267),
|
2441 |
+
Connection(421, 200),
|
2442 |
+
Connection(200, 428),
|
2443 |
+
Connection(428, 421),
|
2444 |
+
Connection(371, 266),
|
2445 |
+
Connection(266, 329),
|
2446 |
+
Connection(329, 371),
|
2447 |
+
Connection(432, 287),
|
2448 |
+
Connection(287, 422),
|
2449 |
+
Connection(422, 432),
|
2450 |
+
Connection(290, 250),
|
2451 |
+
Connection(250, 328),
|
2452 |
+
Connection(328, 290),
|
2453 |
+
Connection(385, 258),
|
2454 |
+
Connection(258, 384),
|
2455 |
+
Connection(384, 385),
|
2456 |
+
Connection(446, 265),
|
2457 |
+
Connection(265, 342),
|
2458 |
+
Connection(342, 446),
|
2459 |
+
Connection(386, 387),
|
2460 |
+
Connection(387, 257),
|
2461 |
+
Connection(257, 386),
|
2462 |
+
Connection(422, 424),
|
2463 |
+
Connection(424, 430),
|
2464 |
+
Connection(430, 422),
|
2465 |
+
Connection(445, 342),
|
2466 |
+
Connection(342, 276),
|
2467 |
+
Connection(276, 445),
|
2468 |
+
Connection(422, 273),
|
2469 |
+
Connection(273, 424),
|
2470 |
+
Connection(424, 422),
|
2471 |
+
Connection(306, 292),
|
2472 |
+
Connection(292, 307),
|
2473 |
+
Connection(307, 306),
|
2474 |
+
Connection(352, 366),
|
2475 |
+
Connection(366, 345),
|
2476 |
+
Connection(345, 352),
|
2477 |
+
Connection(268, 271),
|
2478 |
+
Connection(271, 302),
|
2479 |
+
Connection(302, 268),
|
2480 |
+
Connection(358, 423),
|
2481 |
+
Connection(423, 371),
|
2482 |
+
Connection(371, 358),
|
2483 |
+
Connection(327, 294),
|
2484 |
+
Connection(294, 460),
|
2485 |
+
Connection(460, 327),
|
2486 |
+
Connection(331, 279),
|
2487 |
+
Connection(279, 294),
|
2488 |
+
Connection(294, 331),
|
2489 |
+
Connection(303, 271),
|
2490 |
+
Connection(271, 304),
|
2491 |
+
Connection(304, 303),
|
2492 |
+
Connection(436, 432),
|
2493 |
+
Connection(432, 427),
|
2494 |
+
Connection(427, 436),
|
2495 |
+
Connection(304, 272),
|
2496 |
+
Connection(272, 408),
|
2497 |
+
Connection(408, 304),
|
2498 |
+
Connection(395, 394),
|
2499 |
+
Connection(394, 431),
|
2500 |
+
Connection(431, 395),
|
2501 |
+
Connection(378, 395),
|
2502 |
+
Connection(395, 400),
|
2503 |
+
Connection(400, 378),
|
2504 |
+
Connection(296, 334),
|
2505 |
+
Connection(334, 299),
|
2506 |
+
Connection(299, 296),
|
2507 |
+
Connection(6, 351),
|
2508 |
+
Connection(351, 168),
|
2509 |
+
Connection(168, 6),
|
2510 |
+
Connection(376, 352),
|
2511 |
+
Connection(352, 411),
|
2512 |
+
Connection(411, 376),
|
2513 |
+
Connection(307, 325),
|
2514 |
+
Connection(325, 320),
|
2515 |
+
Connection(320, 307),
|
2516 |
+
Connection(285, 295),
|
2517 |
+
Connection(295, 336),
|
2518 |
+
Connection(336, 285),
|
2519 |
+
Connection(320, 319),
|
2520 |
+
Connection(319, 404),
|
2521 |
+
Connection(404, 320),
|
2522 |
+
Connection(329, 330),
|
2523 |
+
Connection(330, 349),
|
2524 |
+
Connection(349, 329),
|
2525 |
+
Connection(334, 293),
|
2526 |
+
Connection(293, 333),
|
2527 |
+
Connection(333, 334),
|
2528 |
+
Connection(366, 323),
|
2529 |
+
Connection(323, 447),
|
2530 |
+
Connection(447, 366),
|
2531 |
+
Connection(316, 15),
|
2532 |
+
Connection(15, 315),
|
2533 |
+
Connection(315, 316),
|
2534 |
+
Connection(331, 358),
|
2535 |
+
Connection(358, 279),
|
2536 |
+
Connection(279, 331),
|
2537 |
+
Connection(317, 14),
|
2538 |
+
Connection(14, 316),
|
2539 |
+
Connection(316, 317),
|
2540 |
+
Connection(8, 285),
|
2541 |
+
Connection(285, 9),
|
2542 |
+
Connection(9, 8),
|
2543 |
+
Connection(277, 329),
|
2544 |
+
Connection(329, 350),
|
2545 |
+
Connection(350, 277),
|
2546 |
+
Connection(253, 374),
|
2547 |
+
Connection(374, 252),
|
2548 |
+
Connection(252, 253),
|
2549 |
+
Connection(319, 318),
|
2550 |
+
Connection(318, 403),
|
2551 |
+
Connection(403, 319),
|
2552 |
+
Connection(351, 6),
|
2553 |
+
Connection(6, 419),
|
2554 |
+
Connection(419, 351),
|
2555 |
+
Connection(324, 318),
|
2556 |
+
Connection(318, 325),
|
2557 |
+
Connection(325, 324),
|
2558 |
+
Connection(397, 367),
|
2559 |
+
Connection(367, 365),
|
2560 |
+
Connection(365, 397),
|
2561 |
+
Connection(288, 435),
|
2562 |
+
Connection(435, 397),
|
2563 |
+
Connection(397, 288),
|
2564 |
+
Connection(278, 344),
|
2565 |
+
Connection(344, 439),
|
2566 |
+
Connection(439, 278),
|
2567 |
+
Connection(310, 272),
|
2568 |
+
Connection(272, 311),
|
2569 |
+
Connection(311, 310),
|
2570 |
+
Connection(248, 195),
|
2571 |
+
Connection(195, 281),
|
2572 |
+
Connection(281, 248),
|
2573 |
+
Connection(375, 273),
|
2574 |
+
Connection(273, 291),
|
2575 |
+
Connection(291, 375),
|
2576 |
+
Connection(175, 396),
|
2577 |
+
Connection(396, 199),
|
2578 |
+
Connection(199, 175),
|
2579 |
+
Connection(312, 311),
|
2580 |
+
Connection(311, 268),
|
2581 |
+
Connection(268, 312),
|
2582 |
+
Connection(276, 283),
|
2583 |
+
Connection(283, 445),
|
2584 |
+
Connection(445, 276),
|
2585 |
+
Connection(390, 373),
|
2586 |
+
Connection(373, 339),
|
2587 |
+
Connection(339, 390),
|
2588 |
+
Connection(295, 282),
|
2589 |
+
Connection(282, 296),
|
2590 |
+
Connection(296, 295),
|
2591 |
+
Connection(448, 449),
|
2592 |
+
Connection(449, 346),
|
2593 |
+
Connection(346, 448),
|
2594 |
+
Connection(356, 264),
|
2595 |
+
Connection(264, 454),
|
2596 |
+
Connection(454, 356),
|
2597 |
+
Connection(337, 336),
|
2598 |
+
Connection(336, 299),
|
2599 |
+
Connection(299, 337),
|
2600 |
+
Connection(337, 338),
|
2601 |
+
Connection(338, 151),
|
2602 |
+
Connection(151, 337),
|
2603 |
+
Connection(294, 278),
|
2604 |
+
Connection(278, 455),
|
2605 |
+
Connection(455, 294),
|
2606 |
+
Connection(308, 292),
|
2607 |
+
Connection(292, 415),
|
2608 |
+
Connection(415, 308),
|
2609 |
+
Connection(429, 358),
|
2610 |
+
Connection(358, 355),
|
2611 |
+
Connection(355, 429),
|
2612 |
+
Connection(265, 340),
|
2613 |
+
Connection(340, 372),
|
2614 |
+
Connection(372, 265),
|
2615 |
+
Connection(352, 346),
|
2616 |
+
Connection(346, 280),
|
2617 |
+
Connection(280, 352),
|
2618 |
+
Connection(295, 442),
|
2619 |
+
Connection(442, 282),
|
2620 |
+
Connection(282, 295),
|
2621 |
+
Connection(354, 19),
|
2622 |
+
Connection(19, 370),
|
2623 |
+
Connection(370, 354),
|
2624 |
+
Connection(285, 441),
|
2625 |
+
Connection(441, 295),
|
2626 |
+
Connection(295, 285),
|
2627 |
+
Connection(195, 248),
|
2628 |
+
Connection(248, 197),
|
2629 |
+
Connection(197, 195),
|
2630 |
+
Connection(457, 440),
|
2631 |
+
Connection(440, 274),
|
2632 |
+
Connection(274, 457),
|
2633 |
+
Connection(301, 300),
|
2634 |
+
Connection(300, 368),
|
2635 |
+
Connection(368, 301),
|
2636 |
+
Connection(417, 351),
|
2637 |
+
Connection(351, 465),
|
2638 |
+
Connection(465, 417),
|
2639 |
+
Connection(251, 301),
|
2640 |
+
Connection(301, 389),
|
2641 |
+
Connection(389, 251),
|
2642 |
+
Connection(394, 395),
|
2643 |
+
Connection(395, 379),
|
2644 |
+
Connection(379, 394),
|
2645 |
+
Connection(399, 412),
|
2646 |
+
Connection(412, 419),
|
2647 |
+
Connection(419, 399),
|
2648 |
+
Connection(410, 436),
|
2649 |
+
Connection(436, 322),
|
2650 |
+
Connection(322, 410),
|
2651 |
+
Connection(326, 2),
|
2652 |
+
Connection(2, 393),
|
2653 |
+
Connection(393, 326),
|
2654 |
+
Connection(354, 370),
|
2655 |
+
Connection(370, 461),
|
2656 |
+
Connection(461, 354),
|
2657 |
+
Connection(393, 164),
|
2658 |
+
Connection(164, 267),
|
2659 |
+
Connection(267, 393),
|
2660 |
+
Connection(268, 302),
|
2661 |
+
Connection(302, 12),
|
2662 |
+
Connection(12, 268),
|
2663 |
+
Connection(312, 268),
|
2664 |
+
Connection(268, 13),
|
2665 |
+
Connection(13, 312),
|
2666 |
+
Connection(298, 293),
|
2667 |
+
Connection(293, 301),
|
2668 |
+
Connection(301, 298),
|
2669 |
+
Connection(265, 446),
|
2670 |
+
Connection(446, 340),
|
2671 |
+
Connection(340, 265),
|
2672 |
+
Connection(280, 330),
|
2673 |
+
Connection(330, 425),
|
2674 |
+
Connection(425, 280),
|
2675 |
+
Connection(322, 426),
|
2676 |
+
Connection(426, 391),
|
2677 |
+
Connection(391, 322),
|
2678 |
+
Connection(420, 429),
|
2679 |
+
Connection(429, 437),
|
2680 |
+
Connection(437, 420),
|
2681 |
+
Connection(393, 391),
|
2682 |
+
Connection(391, 326),
|
2683 |
+
Connection(326, 393),
|
2684 |
+
Connection(344, 440),
|
2685 |
+
Connection(440, 438),
|
2686 |
+
Connection(438, 344),
|
2687 |
+
Connection(458, 459),
|
2688 |
+
Connection(459, 461),
|
2689 |
+
Connection(461, 458),
|
2690 |
+
Connection(364, 434),
|
2691 |
+
Connection(434, 394),
|
2692 |
+
Connection(394, 364),
|
2693 |
+
Connection(428, 396),
|
2694 |
+
Connection(396, 262),
|
2695 |
+
Connection(262, 428),
|
2696 |
+
Connection(274, 354),
|
2697 |
+
Connection(354, 457),
|
2698 |
+
Connection(457, 274),
|
2699 |
+
Connection(317, 316),
|
2700 |
+
Connection(316, 402),
|
2701 |
+
Connection(402, 317),
|
2702 |
+
Connection(316, 315),
|
2703 |
+
Connection(315, 403),
|
2704 |
+
Connection(403, 316),
|
2705 |
+
Connection(315, 314),
|
2706 |
+
Connection(314, 404),
|
2707 |
+
Connection(404, 315),
|
2708 |
+
Connection(314, 313),
|
2709 |
+
Connection(313, 405),
|
2710 |
+
Connection(405, 314),
|
2711 |
+
Connection(313, 421),
|
2712 |
+
Connection(421, 406),
|
2713 |
+
Connection(406, 313),
|
2714 |
+
Connection(323, 366),
|
2715 |
+
Connection(366, 361),
|
2716 |
+
Connection(361, 323),
|
2717 |
+
Connection(292, 306),
|
2718 |
+
Connection(306, 407),
|
2719 |
+
Connection(407, 292),
|
2720 |
+
Connection(306, 291),
|
2721 |
+
Connection(291, 408),
|
2722 |
+
Connection(408, 306),
|
2723 |
+
Connection(291, 287),
|
2724 |
+
Connection(287, 409),
|
2725 |
+
Connection(409, 291),
|
2726 |
+
Connection(287, 432),
|
2727 |
+
Connection(432, 410),
|
2728 |
+
Connection(410, 287),
|
2729 |
+
Connection(427, 434),
|
2730 |
+
Connection(434, 411),
|
2731 |
+
Connection(411, 427),
|
2732 |
+
Connection(372, 264),
|
2733 |
+
Connection(264, 383),
|
2734 |
+
Connection(383, 372),
|
2735 |
+
Connection(459, 309),
|
2736 |
+
Connection(309, 457),
|
2737 |
+
Connection(457, 459),
|
2738 |
+
Connection(366, 352),
|
2739 |
+
Connection(352, 401),
|
2740 |
+
Connection(401, 366),
|
2741 |
+
Connection(1, 274),
|
2742 |
+
Connection(274, 4),
|
2743 |
+
Connection(4, 1),
|
2744 |
+
Connection(418, 421),
|
2745 |
+
Connection(421, 262),
|
2746 |
+
Connection(262, 418),
|
2747 |
+
Connection(331, 294),
|
2748 |
+
Connection(294, 358),
|
2749 |
+
Connection(358, 331),
|
2750 |
+
Connection(435, 433),
|
2751 |
+
Connection(433, 367),
|
2752 |
+
Connection(367, 435),
|
2753 |
+
Connection(392, 289),
|
2754 |
+
Connection(289, 439),
|
2755 |
+
Connection(439, 392),
|
2756 |
+
Connection(328, 462),
|
2757 |
+
Connection(462, 326),
|
2758 |
+
Connection(326, 328),
|
2759 |
+
Connection(94, 2),
|
2760 |
+
Connection(2, 370),
|
2761 |
+
Connection(370, 94),
|
2762 |
+
Connection(289, 305),
|
2763 |
+
Connection(305, 455),
|
2764 |
+
Connection(455, 289),
|
2765 |
+
Connection(339, 254),
|
2766 |
+
Connection(254, 448),
|
2767 |
+
Connection(448, 339),
|
2768 |
+
Connection(359, 255),
|
2769 |
+
Connection(255, 446),
|
2770 |
+
Connection(446, 359),
|
2771 |
+
Connection(254, 253),
|
2772 |
+
Connection(253, 449),
|
2773 |
+
Connection(449, 254),
|
2774 |
+
Connection(253, 252),
|
2775 |
+
Connection(252, 450),
|
2776 |
+
Connection(450, 253),
|
2777 |
+
Connection(252, 256),
|
2778 |
+
Connection(256, 451),
|
2779 |
+
Connection(451, 252),
|
2780 |
+
Connection(256, 341),
|
2781 |
+
Connection(341, 452),
|
2782 |
+
Connection(452, 256),
|
2783 |
+
Connection(414, 413),
|
2784 |
+
Connection(413, 463),
|
2785 |
+
Connection(463, 414),
|
2786 |
+
Connection(286, 441),
|
2787 |
+
Connection(441, 414),
|
2788 |
+
Connection(414, 286),
|
2789 |
+
Connection(286, 258),
|
2790 |
+
Connection(258, 441),
|
2791 |
+
Connection(441, 286),
|
2792 |
+
Connection(258, 257),
|
2793 |
+
Connection(257, 442),
|
2794 |
+
Connection(442, 258),
|
2795 |
+
Connection(257, 259),
|
2796 |
+
Connection(259, 443),
|
2797 |
+
Connection(443, 257),
|
2798 |
+
Connection(259, 260),
|
2799 |
+
Connection(260, 444),
|
2800 |
+
Connection(444, 259),
|
2801 |
+
Connection(260, 467),
|
2802 |
+
Connection(467, 445),
|
2803 |
+
Connection(445, 260),
|
2804 |
+
Connection(309, 459),
|
2805 |
+
Connection(459, 250),
|
2806 |
+
Connection(250, 309),
|
2807 |
+
Connection(305, 289),
|
2808 |
+
Connection(289, 290),
|
2809 |
+
Connection(290, 305),
|
2810 |
+
Connection(305, 290),
|
2811 |
+
Connection(290, 460),
|
2812 |
+
Connection(460, 305),
|
2813 |
+
Connection(401, 376),
|
2814 |
+
Connection(376, 435),
|
2815 |
+
Connection(435, 401),
|
2816 |
+
Connection(309, 250),
|
2817 |
+
Connection(250, 392),
|
2818 |
+
Connection(392, 309),
|
2819 |
+
Connection(376, 411),
|
2820 |
+
Connection(411, 433),
|
2821 |
+
Connection(433, 376),
|
2822 |
+
Connection(453, 341),
|
2823 |
+
Connection(341, 464),
|
2824 |
+
Connection(464, 453),
|
2825 |
+
Connection(357, 453),
|
2826 |
+
Connection(453, 465),
|
2827 |
+
Connection(465, 357),
|
2828 |
+
Connection(343, 357),
|
2829 |
+
Connection(357, 412),
|
2830 |
+
Connection(412, 343),
|
2831 |
+
Connection(437, 343),
|
2832 |
+
Connection(343, 399),
|
2833 |
+
Connection(399, 437),
|
2834 |
+
Connection(344, 360),
|
2835 |
+
Connection(360, 440),
|
2836 |
+
Connection(440, 344),
|
2837 |
+
Connection(420, 437),
|
2838 |
+
Connection(437, 456),
|
2839 |
+
Connection(456, 420),
|
2840 |
+
Connection(360, 420),
|
2841 |
+
Connection(420, 363),
|
2842 |
+
Connection(363, 360),
|
2843 |
+
Connection(361, 401),
|
2844 |
+
Connection(401, 288),
|
2845 |
+
Connection(288, 361),
|
2846 |
+
Connection(265, 372),
|
2847 |
+
Connection(372, 353),
|
2848 |
+
Connection(353, 265),
|
2849 |
+
Connection(390, 339),
|
2850 |
+
Connection(339, 249),
|
2851 |
+
Connection(249, 390),
|
2852 |
+
Connection(339, 448),
|
2853 |
+
Connection(448, 255),
|
2854 |
+
Connection(255, 339),
|
2855 |
+
]
|
2856 |
+
|
2857 |
+
|
2858 |
+
@dataclasses.dataclass
|
2859 |
+
class FaceLandmarkerResult:
|
2860 |
+
"""The face landmarks detection result from FaceLandmarker, where each vector element represents a single face detected in the image.
|
2861 |
+
|
2862 |
+
Attributes:
|
2863 |
+
face_landmarks: Detected face landmarks in normalized image coordinates.
|
2864 |
+
face_blendshapes: Optional face blendshapes results.
|
2865 |
+
facial_transformation_matrixes: Optional facial transformation matrix.
|
2866 |
+
"""
|
2867 |
+
|
2868 |
+
face_landmarks: List[List[landmark_module.NormalizedLandmark]]
|
2869 |
+
face_blendshapes: List[List[category_module.Category]]
|
2870 |
+
facial_transformation_matrixes: List[np.ndarray]
|
2871 |
+
|
2872 |
+
|
2873 |
+
def _build_landmarker_result(
|
2874 |
+
output_packets: Mapping[str, packet_module.Packet]
|
2875 |
+
) -> FaceLandmarkerResult:
|
2876 |
+
"""Constructs a `FaceLandmarkerResult` from output packets."""
|
2877 |
+
face_landmarks_proto_list = packet_getter.get_proto_list(
|
2878 |
+
output_packets[_NORM_LANDMARKS_STREAM_NAME]
|
2879 |
+
)
|
2880 |
+
|
2881 |
+
face_landmarks_results = []
|
2882 |
+
for proto in face_landmarks_proto_list:
|
2883 |
+
face_landmarks = landmark_pb2.NormalizedLandmarkList()
|
2884 |
+
face_landmarks.MergeFrom(proto)
|
2885 |
+
face_landmarks_list = []
|
2886 |
+
for face_landmark in face_landmarks.landmark:
|
2887 |
+
face_landmarks_list.append(
|
2888 |
+
landmark_module.NormalizedLandmark.create_from_pb2(face_landmark)
|
2889 |
+
)
|
2890 |
+
face_landmarks_results.append(face_landmarks_list)
|
2891 |
+
|
2892 |
+
face_blendshapes_results = []
|
2893 |
+
if _BLENDSHAPES_STREAM_NAME in output_packets:
|
2894 |
+
face_blendshapes_proto_list = packet_getter.get_proto_list(
|
2895 |
+
output_packets[_BLENDSHAPES_STREAM_NAME]
|
2896 |
+
)
|
2897 |
+
for proto in face_blendshapes_proto_list:
|
2898 |
+
face_blendshapes_categories = []
|
2899 |
+
face_blendshapes_classifications = classification_pb2.ClassificationList()
|
2900 |
+
face_blendshapes_classifications.MergeFrom(proto)
|
2901 |
+
for face_blendshapes in face_blendshapes_classifications.classification:
|
2902 |
+
face_blendshapes_categories.append(
|
2903 |
+
category_module.Category(
|
2904 |
+
index=face_blendshapes.index,
|
2905 |
+
score=face_blendshapes.score,
|
2906 |
+
display_name=face_blendshapes.display_name,
|
2907 |
+
category_name=face_blendshapes.label,
|
2908 |
+
)
|
2909 |
+
)
|
2910 |
+
face_blendshapes_results.append(face_blendshapes_categories)
|
2911 |
+
|
2912 |
+
facial_transformation_matrixes_results = []
|
2913 |
+
if _FACE_GEOMETRY_STREAM_NAME in output_packets:
|
2914 |
+
facial_transformation_matrixes_proto_list = packet_getter.get_proto_list(
|
2915 |
+
output_packets[_FACE_GEOMETRY_STREAM_NAME]
|
2916 |
+
)
|
2917 |
+
for proto in facial_transformation_matrixes_proto_list:
|
2918 |
+
if hasattr(proto, 'pose_transform_matrix'):
|
2919 |
+
matrix_data = matrix_data_pb2.MatrixData()
|
2920 |
+
matrix_data.MergeFrom(proto.pose_transform_matrix)
|
2921 |
+
matrix = np.array(matrix_data.packed_data)
|
2922 |
+
matrix = matrix.reshape((matrix_data.rows, matrix_data.cols))
|
2923 |
+
matrix = (
|
2924 |
+
matrix if matrix_data.layout == _LayoutEnum.ROW_MAJOR else matrix.T
|
2925 |
+
)
|
2926 |
+
facial_transformation_matrixes_results.append(matrix)
|
2927 |
+
|
2928 |
+
return FaceLandmarkerResult(
|
2929 |
+
face_landmarks_results,
|
2930 |
+
face_blendshapes_results,
|
2931 |
+
facial_transformation_matrixes_results,
|
2932 |
+
)
|
2933 |
+
|
2934 |
+
def _build_landmarker_result2(
|
2935 |
+
output_packets: Mapping[str, packet_module.Packet]
|
2936 |
+
) -> FaceLandmarkerResult:
|
2937 |
+
"""Constructs a `FaceLandmarkerResult` from output packets."""
|
2938 |
+
face_landmarks_proto_list = packet_getter.get_proto_list(
|
2939 |
+
output_packets[_NORM_LANDMARKS_STREAM_NAME]
|
2940 |
+
)
|
2941 |
+
|
2942 |
+
face_landmarks_results = []
|
2943 |
+
for proto in face_landmarks_proto_list:
|
2944 |
+
face_landmarks = landmark_pb2.NormalizedLandmarkList()
|
2945 |
+
face_landmarks.MergeFrom(proto)
|
2946 |
+
face_landmarks_list = []
|
2947 |
+
for face_landmark in face_landmarks.landmark:
|
2948 |
+
face_landmarks_list.append(
|
2949 |
+
landmark_module.NormalizedLandmark.create_from_pb2(face_landmark)
|
2950 |
+
)
|
2951 |
+
face_landmarks_results.append(face_landmarks_list)
|
2952 |
+
|
2953 |
+
face_blendshapes_results = []
|
2954 |
+
if _BLENDSHAPES_STREAM_NAME in output_packets:
|
2955 |
+
face_blendshapes_proto_list = packet_getter.get_proto_list(
|
2956 |
+
output_packets[_BLENDSHAPES_STREAM_NAME]
|
2957 |
+
)
|
2958 |
+
for proto in face_blendshapes_proto_list:
|
2959 |
+
face_blendshapes_categories = []
|
2960 |
+
face_blendshapes_classifications = classification_pb2.ClassificationList()
|
2961 |
+
face_blendshapes_classifications.MergeFrom(proto)
|
2962 |
+
for face_blendshapes in face_blendshapes_classifications.classification:
|
2963 |
+
face_blendshapes_categories.append(
|
2964 |
+
category_module.Category(
|
2965 |
+
index=face_blendshapes.index,
|
2966 |
+
score=face_blendshapes.score,
|
2967 |
+
display_name=face_blendshapes.display_name,
|
2968 |
+
category_name=face_blendshapes.label,
|
2969 |
+
)
|
2970 |
+
)
|
2971 |
+
face_blendshapes_results.append(face_blendshapes_categories)
|
2972 |
+
|
2973 |
+
facial_transformation_matrixes_results = []
|
2974 |
+
if _FACE_GEOMETRY_STREAM_NAME in output_packets:
|
2975 |
+
facial_transformation_matrixes_proto_list = packet_getter.get_proto_list(
|
2976 |
+
output_packets[_FACE_GEOMETRY_STREAM_NAME]
|
2977 |
+
)
|
2978 |
+
for proto in facial_transformation_matrixes_proto_list:
|
2979 |
+
if hasattr(proto, 'pose_transform_matrix'):
|
2980 |
+
matrix_data = matrix_data_pb2.MatrixData()
|
2981 |
+
matrix_data.MergeFrom(proto.pose_transform_matrix)
|
2982 |
+
matrix = np.array(matrix_data.packed_data)
|
2983 |
+
matrix = matrix.reshape((matrix_data.rows, matrix_data.cols))
|
2984 |
+
matrix = (
|
2985 |
+
matrix if matrix_data.layout == _LayoutEnum.ROW_MAJOR else matrix.T
|
2986 |
+
)
|
2987 |
+
facial_transformation_matrixes_results.append(matrix)
|
2988 |
+
|
2989 |
+
return FaceLandmarkerResult(
|
2990 |
+
face_landmarks_results,
|
2991 |
+
face_blendshapes_results,
|
2992 |
+
facial_transformation_matrixes_results,
|
2993 |
+
), facial_transformation_matrixes_proto_list[0].mesh
|
2994 |
+
|
2995 |
+
@dataclasses.dataclass
|
2996 |
+
class FaceLandmarkerOptions:
|
2997 |
+
"""Options for the face landmarker task.
|
2998 |
+
|
2999 |
+
Attributes:
|
3000 |
+
base_options: Base options for the face landmarker task.
|
3001 |
+
running_mode: The running mode of the task. Default to the image mode.
|
3002 |
+
FaceLandmarker has three running modes: 1) The image mode for detecting
|
3003 |
+
face landmarks on single image inputs. 2) The video mode for detecting
|
3004 |
+
face landmarks on the decoded frames of a video. 3) The live stream mode
|
3005 |
+
for detecting face landmarks on the live stream of input data, such as
|
3006 |
+
from camera. In this mode, the "result_callback" below must be specified
|
3007 |
+
to receive the detection results asynchronously.
|
3008 |
+
num_faces: The maximum number of faces that can be detected by the
|
3009 |
+
FaceLandmarker.
|
3010 |
+
min_face_detection_confidence: The minimum confidence score for the face
|
3011 |
+
detection to be considered successful.
|
3012 |
+
min_face_presence_confidence: The minimum confidence score of face presence
|
3013 |
+
score in the face landmark detection.
|
3014 |
+
min_tracking_confidence: The minimum confidence score for the face tracking
|
3015 |
+
to be considered successful.
|
3016 |
+
output_face_blendshapes: Whether FaceLandmarker outputs face blendshapes
|
3017 |
+
classification. Face blendshapes are used for rendering the 3D face model.
|
3018 |
+
output_facial_transformation_matrixes: Whether FaceLandmarker outputs facial
|
3019 |
+
transformation_matrix. Facial transformation matrix is used to transform
|
3020 |
+
the face landmarks in canonical face to the detected face, so that users
|
3021 |
+
can apply face effects on the detected landmarks.
|
3022 |
+
result_callback: The user-defined result callback for processing live stream
|
3023 |
+
data. The result callback should only be specified when the running mode
|
3024 |
+
is set to the live stream mode.
|
3025 |
+
"""
|
3026 |
+
|
3027 |
+
base_options: _BaseOptions
|
3028 |
+
running_mode: _RunningMode = _RunningMode.IMAGE
|
3029 |
+
num_faces: int = 1
|
3030 |
+
min_face_detection_confidence: float = 0.5
|
3031 |
+
min_face_presence_confidence: float = 0.5
|
3032 |
+
min_tracking_confidence: float = 0.5
|
3033 |
+
output_face_blendshapes: bool = False
|
3034 |
+
output_facial_transformation_matrixes: bool = False
|
3035 |
+
result_callback: Optional[
|
3036 |
+
Callable[[FaceLandmarkerResult, image_module.Image, int], None]
|
3037 |
+
] = None
|
3038 |
+
|
3039 |
+
@doc_controls.do_not_generate_docs
|
3040 |
+
def to_pb2(self) -> _FaceLandmarkerGraphOptionsProto:
|
3041 |
+
"""Generates an FaceLandmarkerGraphOptions protobuf object."""
|
3042 |
+
base_options_proto = self.base_options.to_pb2()
|
3043 |
+
base_options_proto.use_stream_mode = (
|
3044 |
+
False if self.running_mode == _RunningMode.IMAGE else True
|
3045 |
+
)
|
3046 |
+
|
3047 |
+
# Initialize the face landmarker options from base options.
|
3048 |
+
face_landmarker_options_proto = _FaceLandmarkerGraphOptionsProto(
|
3049 |
+
base_options=base_options_proto
|
3050 |
+
)
|
3051 |
+
|
3052 |
+
# Configure face detector options.
|
3053 |
+
face_landmarker_options_proto.face_detector_graph_options.num_faces = (
|
3054 |
+
self.num_faces
|
3055 |
+
)
|
3056 |
+
face_landmarker_options_proto.face_detector_graph_options.min_detection_confidence = (
|
3057 |
+
self.min_face_detection_confidence
|
3058 |
+
)
|
3059 |
+
|
3060 |
+
# Configure face landmark detector options.
|
3061 |
+
face_landmarker_options_proto.min_tracking_confidence = (
|
3062 |
+
self.min_tracking_confidence
|
3063 |
+
)
|
3064 |
+
face_landmarker_options_proto.face_landmarks_detector_graph_options.min_detection_confidence = (
|
3065 |
+
self.min_face_detection_confidence
|
3066 |
+
)
|
3067 |
+
return face_landmarker_options_proto
|
3068 |
+
|
3069 |
+
|
3070 |
+
class FaceLandmarker(base_vision_task_api.BaseVisionTaskApi):
|
3071 |
+
"""Class that performs face landmarks detection on images."""
|
3072 |
+
|
3073 |
+
@classmethod
|
3074 |
+
def create_from_model_path(cls, model_path: str) -> 'FaceLandmarker':
|
3075 |
+
"""Creates an `FaceLandmarker` object from a TensorFlow Lite model and the default `FaceLandmarkerOptions`.
|
3076 |
+
|
3077 |
+
Note that the created `FaceLandmarker` instance is in image mode, for
|
3078 |
+
detecting face landmarks on single image inputs.
|
3079 |
+
|
3080 |
+
Args:
|
3081 |
+
model_path: Path to the model.
|
3082 |
+
|
3083 |
+
Returns:
|
3084 |
+
`FaceLandmarker` object that's created from the model file and the
|
3085 |
+
default `FaceLandmarkerOptions`.
|
3086 |
+
|
3087 |
+
Raises:
|
3088 |
+
ValueError: If failed to create `FaceLandmarker` object from the
|
3089 |
+
provided file such as invalid file path.
|
3090 |
+
RuntimeError: If other types of error occurred.
|
3091 |
+
"""
|
3092 |
+
base_options = _BaseOptions(model_asset_path=model_path)
|
3093 |
+
options = FaceLandmarkerOptions(
|
3094 |
+
base_options=base_options, running_mode=_RunningMode.IMAGE
|
3095 |
+
)
|
3096 |
+
return cls.create_from_options(options)
|
3097 |
+
|
3098 |
+
@classmethod
|
3099 |
+
def create_from_options(
|
3100 |
+
cls, options: FaceLandmarkerOptions
|
3101 |
+
) -> 'FaceLandmarker':
|
3102 |
+
"""Creates the `FaceLandmarker` object from face landmarker options.
|
3103 |
+
|
3104 |
+
Args:
|
3105 |
+
options: Options for the face landmarker task.
|
3106 |
+
|
3107 |
+
Returns:
|
3108 |
+
`FaceLandmarker` object that's created from `options`.
|
3109 |
+
|
3110 |
+
Raises:
|
3111 |
+
ValueError: If failed to create `FaceLandmarker` object from
|
3112 |
+
`FaceLandmarkerOptions` such as missing the model.
|
3113 |
+
RuntimeError: If other types of error occurred.
|
3114 |
+
"""
|
3115 |
+
|
3116 |
+
def packets_callback(output_packets: Mapping[str, packet_module.Packet]):
|
3117 |
+
if output_packets[_IMAGE_OUT_STREAM_NAME].is_empty():
|
3118 |
+
return
|
3119 |
+
|
3120 |
+
image = packet_getter.get_image(output_packets[_IMAGE_OUT_STREAM_NAME])
|
3121 |
+
if output_packets[_IMAGE_OUT_STREAM_NAME].is_empty():
|
3122 |
+
return
|
3123 |
+
|
3124 |
+
if output_packets[_NORM_LANDMARKS_STREAM_NAME].is_empty():
|
3125 |
+
empty_packet = output_packets[_NORM_LANDMARKS_STREAM_NAME]
|
3126 |
+
options.result_callback(
|
3127 |
+
FaceLandmarkerResult([], [], []),
|
3128 |
+
image,
|
3129 |
+
empty_packet.timestamp.value // _MICRO_SECONDS_PER_MILLISECOND,
|
3130 |
+
)
|
3131 |
+
return
|
3132 |
+
|
3133 |
+
face_landmarks_result = _build_landmarker_result(output_packets)
|
3134 |
+
timestamp = output_packets[_NORM_LANDMARKS_STREAM_NAME].timestamp
|
3135 |
+
options.result_callback(
|
3136 |
+
face_landmarks_result,
|
3137 |
+
image,
|
3138 |
+
timestamp.value // _MICRO_SECONDS_PER_MILLISECOND,
|
3139 |
+
)
|
3140 |
+
|
3141 |
+
output_streams = [
|
3142 |
+
':'.join([_NORM_LANDMARKS_TAG, _NORM_LANDMARKS_STREAM_NAME]),
|
3143 |
+
':'.join([_IMAGE_TAG, _IMAGE_OUT_STREAM_NAME]),
|
3144 |
+
]
|
3145 |
+
|
3146 |
+
if options.output_face_blendshapes:
|
3147 |
+
output_streams.append(
|
3148 |
+
':'.join([_BLENDSHAPES_TAG, _BLENDSHAPES_STREAM_NAME])
|
3149 |
+
)
|
3150 |
+
if options.output_facial_transformation_matrixes:
|
3151 |
+
output_streams.append(
|
3152 |
+
':'.join([_FACE_GEOMETRY_TAG, _FACE_GEOMETRY_STREAM_NAME])
|
3153 |
+
)
|
3154 |
+
|
3155 |
+
task_info = _TaskInfo(
|
3156 |
+
task_graph=_TASK_GRAPH_NAME,
|
3157 |
+
input_streams=[
|
3158 |
+
':'.join([_IMAGE_TAG, _IMAGE_IN_STREAM_NAME]),
|
3159 |
+
':'.join([_NORM_RECT_TAG, _NORM_RECT_STREAM_NAME]),
|
3160 |
+
],
|
3161 |
+
output_streams=output_streams,
|
3162 |
+
task_options=options,
|
3163 |
+
)
|
3164 |
+
return cls(
|
3165 |
+
task_info.generate_graph_config(
|
3166 |
+
enable_flow_limiting=options.running_mode
|
3167 |
+
== _RunningMode.LIVE_STREAM
|
3168 |
+
),
|
3169 |
+
options.running_mode,
|
3170 |
+
packets_callback if options.result_callback else None,
|
3171 |
+
)
|
3172 |
+
|
3173 |
+
def detect(
|
3174 |
+
self,
|
3175 |
+
image: image_module.Image,
|
3176 |
+
image_processing_options: Optional[_ImageProcessingOptions] = None,
|
3177 |
+
) -> FaceLandmarkerResult:
|
3178 |
+
"""Performs face landmarks detection on the given image.
|
3179 |
+
|
3180 |
+
Only use this method when the FaceLandmarker is created with the image
|
3181 |
+
running mode.
|
3182 |
+
|
3183 |
+
The image can be of any size with format RGB or RGBA.
|
3184 |
+
TODO: Describes how the input image will be preprocessed after the yuv
|
3185 |
+
support is implemented.
|
3186 |
+
|
3187 |
+
Args:
|
3188 |
+
image: MediaPipe Image.
|
3189 |
+
image_processing_options: Options for image processing.
|
3190 |
+
|
3191 |
+
Returns:
|
3192 |
+
The face landmarks detection results.
|
3193 |
+
|
3194 |
+
Raises:
|
3195 |
+
ValueError: If any of the input arguments is invalid.
|
3196 |
+
RuntimeError: If face landmarker detection failed to run.
|
3197 |
+
"""
|
3198 |
+
|
3199 |
+
normalized_rect = self.convert_to_normalized_rect(
|
3200 |
+
image_processing_options, image, roi_allowed=False
|
3201 |
+
)
|
3202 |
+
output_packets = self._process_image_data({
|
3203 |
+
_IMAGE_IN_STREAM_NAME: packet_creator.create_image(image),
|
3204 |
+
_NORM_RECT_STREAM_NAME: packet_creator.create_proto(
|
3205 |
+
normalized_rect.to_pb2()
|
3206 |
+
),
|
3207 |
+
})
|
3208 |
+
|
3209 |
+
if output_packets[_NORM_LANDMARKS_STREAM_NAME].is_empty():
|
3210 |
+
return FaceLandmarkerResult([], [], [])
|
3211 |
+
|
3212 |
+
return _build_landmarker_result2(output_packets)
|
3213 |
+
|
3214 |
+
def detect_for_video(
|
3215 |
+
self,
|
3216 |
+
image: image_module.Image,
|
3217 |
+
timestamp_ms: int,
|
3218 |
+
image_processing_options: Optional[_ImageProcessingOptions] = None,
|
3219 |
+
):
|
3220 |
+
"""Performs face landmarks detection on the provided video frame.
|
3221 |
+
|
3222 |
+
Only use this method when the FaceLandmarker is created with the video
|
3223 |
+
running mode.
|
3224 |
+
|
3225 |
+
Only use this method when the FaceLandmarker is created with the video
|
3226 |
+
running mode. It's required to provide the video frame's timestamp (in
|
3227 |
+
milliseconds) along with the video frame. The input timestamps should be
|
3228 |
+
monotonically increasing for adjacent calls of this method.
|
3229 |
+
|
3230 |
+
Args:
|
3231 |
+
image: MediaPipe Image.
|
3232 |
+
timestamp_ms: The timestamp of the input video frame in milliseconds.
|
3233 |
+
image_processing_options: Options for image processing.
|
3234 |
+
|
3235 |
+
Returns:
|
3236 |
+
The face landmarks detection results.
|
3237 |
+
|
3238 |
+
Raises:
|
3239 |
+
ValueError: If any of the input arguments is invalid.
|
3240 |
+
RuntimeError: If face landmarker detection failed to run.
|
3241 |
+
"""
|
3242 |
+
normalized_rect = self.convert_to_normalized_rect(
|
3243 |
+
image_processing_options, image, roi_allowed=False
|
3244 |
+
)
|
3245 |
+
output_packets = self._process_video_data({
|
3246 |
+
_IMAGE_IN_STREAM_NAME: packet_creator.create_image(image).at(
|
3247 |
+
timestamp_ms * _MICRO_SECONDS_PER_MILLISECOND
|
3248 |
+
),
|
3249 |
+
_NORM_RECT_STREAM_NAME: packet_creator.create_proto(
|
3250 |
+
normalized_rect.to_pb2()
|
3251 |
+
).at(timestamp_ms * _MICRO_SECONDS_PER_MILLISECOND),
|
3252 |
+
})
|
3253 |
+
|
3254 |
+
if output_packets[_NORM_LANDMARKS_STREAM_NAME].is_empty():
|
3255 |
+
return FaceLandmarkerResult([], [], [])
|
3256 |
+
|
3257 |
+
return _build_landmarker_result2(output_packets)
|
3258 |
+
|
3259 |
+
def detect_async(
|
3260 |
+
self,
|
3261 |
+
image: image_module.Image,
|
3262 |
+
timestamp_ms: int,
|
3263 |
+
image_processing_options: Optional[_ImageProcessingOptions] = None,
|
3264 |
+
) -> None:
|
3265 |
+
"""Sends live image data to perform face landmarks detection.
|
3266 |
+
|
3267 |
+
The results will be available via the "result_callback" provided in the
|
3268 |
+
FaceLandmarkerOptions. Only use this method when the FaceLandmarker is
|
3269 |
+
created with the live stream running mode.
|
3270 |
+
|
3271 |
+
Only use this method when the FaceLandmarker is created with the live
|
3272 |
+
stream running mode. The input timestamps should be monotonically increasing
|
3273 |
+
for adjacent calls of this method. This method will return immediately after
|
3274 |
+
the input image is accepted. The results will be available via the
|
3275 |
+
`result_callback` provided in the `FaceLandmarkerOptions`. The
|
3276 |
+
`detect_async` method is designed to process live stream data such as
|
3277 |
+
camera input. To lower the overall latency, face landmarker may drop the
|
3278 |
+
input images if needed. In other words, it's not guaranteed to have output
|
3279 |
+
per input image.
|
3280 |
+
|
3281 |
+
The `result_callback` provides:
|
3282 |
+
- The face landmarks detection results.
|
3283 |
+
- The input image that the face landmarker runs on.
|
3284 |
+
- The input timestamp in milliseconds.
|
3285 |
+
|
3286 |
+
Args:
|
3287 |
+
image: MediaPipe Image.
|
3288 |
+
timestamp_ms: The timestamp of the input image in milliseconds.
|
3289 |
+
image_processing_options: Options for image processing.
|
3290 |
+
|
3291 |
+
Raises:
|
3292 |
+
ValueError: If the current input timestamp is smaller than what the
|
3293 |
+
face landmarker has already processed.
|
3294 |
+
"""
|
3295 |
+
normalized_rect = self.convert_to_normalized_rect(
|
3296 |
+
image_processing_options, image, roi_allowed=False
|
3297 |
+
)
|
3298 |
+
self._send_live_stream_data({
|
3299 |
+
_IMAGE_IN_STREAM_NAME: packet_creator.create_image(image).at(
|
3300 |
+
timestamp_ms * _MICRO_SECONDS_PER_MILLISECOND
|
3301 |
+
),
|
3302 |
+
_NORM_RECT_STREAM_NAME: packet_creator.create_proto(
|
3303 |
+
normalized_rect.to_pb2()
|
3304 |
+
).at(timestamp_ms * _MICRO_SECONDS_PER_MILLISECOND),
|
3305 |
+
})
|
src/utils/mp_utils.py
ADDED
@@ -0,0 +1,95 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import numpy as np
|
3 |
+
import cv2
|
4 |
+
import time
|
5 |
+
from tqdm import tqdm
|
6 |
+
import multiprocessing
|
7 |
+
import glob
|
8 |
+
|
9 |
+
import mediapipe as mp
|
10 |
+
from mediapipe import solutions
|
11 |
+
from mediapipe.framework.formats import landmark_pb2
|
12 |
+
from mediapipe.tasks import python
|
13 |
+
from mediapipe.tasks.python import vision
|
14 |
+
from . import face_landmark
|
15 |
+
|
16 |
+
CUR_DIR = os.path.dirname(__file__)
|
17 |
+
|
18 |
+
|
19 |
+
class LMKExtractor():
|
20 |
+
def __init__(self, FPS=25):
|
21 |
+
# Create an FaceLandmarker object.
|
22 |
+
self.mode = mp.tasks.vision.FaceDetectorOptions.running_mode.IMAGE
|
23 |
+
base_options = python.BaseOptions(model_asset_path=os.path.join(CUR_DIR, 'mp_models/face_landmarker_v2_with_blendshapes.task'))
|
24 |
+
base_options.delegate = mp.tasks.BaseOptions.Delegate.CPU
|
25 |
+
options = vision.FaceLandmarkerOptions(base_options=base_options,
|
26 |
+
running_mode=self.mode,
|
27 |
+
output_face_blendshapes=True,
|
28 |
+
output_facial_transformation_matrixes=True,
|
29 |
+
num_faces=1)
|
30 |
+
self.detector = face_landmark.FaceLandmarker.create_from_options(options)
|
31 |
+
self.last_ts = 0
|
32 |
+
self.frame_ms = int(1000 / FPS)
|
33 |
+
|
34 |
+
det_base_options = python.BaseOptions(model_asset_path=os.path.join(CUR_DIR, 'mp_models/blaze_face_short_range.tflite'))
|
35 |
+
det_options = vision.FaceDetectorOptions(base_options=det_base_options)
|
36 |
+
self.det_detector = vision.FaceDetector.create_from_options(det_options)
|
37 |
+
|
38 |
+
|
39 |
+
def __call__(self, img):
|
40 |
+
frame = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
|
41 |
+
image = mp.Image(image_format=mp.ImageFormat.SRGB, data=frame)
|
42 |
+
t0 = time.time()
|
43 |
+
if self.mode == mp.tasks.vision.FaceDetectorOptions.running_mode.VIDEO:
|
44 |
+
det_result = self.det_detector.detect(image)
|
45 |
+
if len(det_result.detections) != 1:
|
46 |
+
return None
|
47 |
+
self.last_ts += self.frame_ms
|
48 |
+
try:
|
49 |
+
detection_result, mesh3d = self.detector.detect_for_video(image, timestamp_ms=self.last_ts)
|
50 |
+
except:
|
51 |
+
return None
|
52 |
+
elif self.mode == mp.tasks.vision.FaceDetectorOptions.running_mode.IMAGE:
|
53 |
+
# det_result = self.det_detector.detect(image)
|
54 |
+
|
55 |
+
# if len(det_result.detections) != 1:
|
56 |
+
# return None
|
57 |
+
try:
|
58 |
+
detection_result, mesh3d = self.detector.detect(image)
|
59 |
+
except:
|
60 |
+
return None
|
61 |
+
|
62 |
+
|
63 |
+
bs_list = detection_result.face_blendshapes
|
64 |
+
if len(bs_list) == 1:
|
65 |
+
bs = bs_list[0]
|
66 |
+
bs_values = []
|
67 |
+
for index in range(len(bs)):
|
68 |
+
bs_values.append(bs[index].score)
|
69 |
+
bs_values = bs_values[1:] # remove neutral
|
70 |
+
trans_mat = detection_result.facial_transformation_matrixes[0]
|
71 |
+
face_landmarks_list = detection_result.face_landmarks
|
72 |
+
face_landmarks = face_landmarks_list[0]
|
73 |
+
lmks = []
|
74 |
+
for index in range(len(face_landmarks)):
|
75 |
+
x = face_landmarks[index].x
|
76 |
+
y = face_landmarks[index].y
|
77 |
+
z = face_landmarks[index].z
|
78 |
+
lmks.append([x, y, z])
|
79 |
+
lmks = np.array(lmks)
|
80 |
+
|
81 |
+
lmks3d = np.array(mesh3d.vertex_buffer)
|
82 |
+
lmks3d = lmks3d.reshape(-1, 5)[:, :3]
|
83 |
+
mp_tris = np.array(mesh3d.index_buffer).reshape(-1, 3) + 1
|
84 |
+
|
85 |
+
return {
|
86 |
+
"lmks": lmks,
|
87 |
+
'lmks3d': lmks3d,
|
88 |
+
"trans_mat": trans_mat,
|
89 |
+
'faces': mp_tris,
|
90 |
+
"bs": bs_values
|
91 |
+
}
|
92 |
+
else:
|
93 |
+
# print('multiple faces in the image: {}'.format(img_path))
|
94 |
+
return None
|
95 |
+
|
src/utils/pose_util.py
ADDED
@@ -0,0 +1,78 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import math
|
2 |
+
|
3 |
+
import numpy as np
|
4 |
+
from scipy.spatial.transform import Rotation as R
|
5 |
+
|
6 |
+
|
7 |
+
def create_perspective_matrix(aspect_ratio):
|
8 |
+
kDegreesToRadians = np.pi / 180.
|
9 |
+
near = 1
|
10 |
+
far = 10000
|
11 |
+
perspective_matrix = np.zeros(16, dtype=np.float32)
|
12 |
+
|
13 |
+
# Standard perspective projection matrix calculations.
|
14 |
+
f = 1.0 / np.tan(kDegreesToRadians * 63 / 2.)
|
15 |
+
|
16 |
+
denom = 1.0 / (near - far)
|
17 |
+
perspective_matrix[0] = f / aspect_ratio
|
18 |
+
perspective_matrix[5] = f
|
19 |
+
perspective_matrix[10] = (near + far) * denom
|
20 |
+
perspective_matrix[11] = -1.
|
21 |
+
perspective_matrix[14] = 1. * far * near * denom
|
22 |
+
|
23 |
+
# If the environment's origin point location is in the top left corner,
|
24 |
+
# then skip additional flip along Y-axis is required to render correctly.
|
25 |
+
|
26 |
+
perspective_matrix[5] *= -1.
|
27 |
+
return perspective_matrix
|
28 |
+
|
29 |
+
|
30 |
+
def project_points(points_3d, transformation_matrix, pose_vectors, image_shape):
|
31 |
+
P = create_perspective_matrix(image_shape[1] / image_shape[0]).reshape(4, 4).T
|
32 |
+
L, N, _ = points_3d.shape
|
33 |
+
projected_points = np.zeros((L, N, 2))
|
34 |
+
for i in range(L):
|
35 |
+
points_3d_frame = points_3d[i]
|
36 |
+
ones = np.ones((points_3d_frame.shape[0], 1))
|
37 |
+
points_3d_homogeneous = np.hstack([points_3d_frame, ones])
|
38 |
+
transformed_points = points_3d_homogeneous @ (transformation_matrix @ euler_and_translation_to_matrix(pose_vectors[i][:3], pose_vectors[i][3:])).T @ P
|
39 |
+
projected_points_frame = transformed_points[:, :2] / transformed_points[:, 3, np.newaxis] # -1 ~ 1
|
40 |
+
projected_points_frame[:, 0] = (projected_points_frame[:, 0] + 1) * 0.5 * image_shape[1]
|
41 |
+
projected_points_frame[:, 1] = (projected_points_frame[:, 1] + 1) * 0.5 * image_shape[0]
|
42 |
+
projected_points[i] = projected_points_frame
|
43 |
+
return projected_points
|
44 |
+
|
45 |
+
|
46 |
+
def project_points_with_trans(points_3d, transformation_matrix, image_shape):
|
47 |
+
P = create_perspective_matrix(image_shape[1] / image_shape[0]).reshape(4, 4).T
|
48 |
+
L, N, _ = points_3d.shape
|
49 |
+
projected_points = np.zeros((L, N, 2))
|
50 |
+
for i in range(L):
|
51 |
+
points_3d_frame = points_3d[i]
|
52 |
+
ones = np.ones((points_3d_frame.shape[0], 1))
|
53 |
+
points_3d_homogeneous = np.hstack([points_3d_frame, ones])
|
54 |
+
transformed_points = points_3d_homogeneous @ transformation_matrix[i].T @ P
|
55 |
+
projected_points_frame = transformed_points[:, :2] / transformed_points[:, 3, np.newaxis] # -1 ~ 1
|
56 |
+
projected_points_frame[:, 0] = (projected_points_frame[:, 0] + 1) * 0.5 * image_shape[1]
|
57 |
+
projected_points_frame[:, 1] = (projected_points_frame[:, 1] + 1) * 0.5 * image_shape[0]
|
58 |
+
projected_points[i] = projected_points_frame
|
59 |
+
return projected_points
|
60 |
+
|
61 |
+
|
62 |
+
def euler_and_translation_to_matrix(euler_angles, translation_vector):
|
63 |
+
rotation = R.from_euler('xyz', euler_angles, degrees=True)
|
64 |
+
rotation_matrix = rotation.as_matrix()
|
65 |
+
|
66 |
+
matrix = np.eye(4)
|
67 |
+
matrix[:3, :3] = rotation_matrix
|
68 |
+
matrix[:3, 3] = translation_vector
|
69 |
+
|
70 |
+
return matrix
|
71 |
+
|
72 |
+
|
73 |
+
def matrix_to_euler_and_translation(matrix):
|
74 |
+
rotation_matrix = matrix[:3, :3]
|
75 |
+
translation_vector = matrix[:3, 3]
|
76 |
+
rotation = R.from_matrix(rotation_matrix)
|
77 |
+
euler_angles = rotation.as_euler('xyz', degrees=True)
|
78 |
+
return euler_angles, translation_vector
|
src/utils/util.py
ADDED
@@ -0,0 +1,128 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import importlib
|
2 |
+
import os
|
3 |
+
import os.path as osp
|
4 |
+
import shutil
|
5 |
+
import sys
|
6 |
+
from pathlib import Path
|
7 |
+
|
8 |
+
import av
|
9 |
+
import numpy as np
|
10 |
+
import torch
|
11 |
+
import torchvision
|
12 |
+
from einops import rearrange
|
13 |
+
from PIL import Image
|
14 |
+
|
15 |
+
|
16 |
+
def seed_everything(seed):
|
17 |
+
import random
|
18 |
+
|
19 |
+
import numpy as np
|
20 |
+
|
21 |
+
torch.manual_seed(seed)
|
22 |
+
torch.cuda.manual_seed_all(seed)
|
23 |
+
np.random.seed(seed % (2**32))
|
24 |
+
random.seed(seed)
|
25 |
+
|
26 |
+
|
27 |
+
def import_filename(filename):
|
28 |
+
spec = importlib.util.spec_from_file_location("mymodule", filename)
|
29 |
+
module = importlib.util.module_from_spec(spec)
|
30 |
+
sys.modules[spec.name] = module
|
31 |
+
spec.loader.exec_module(module)
|
32 |
+
return module
|
33 |
+
|
34 |
+
|
35 |
+
def delete_additional_ckpt(base_path, num_keep):
|
36 |
+
dirs = []
|
37 |
+
for d in os.listdir(base_path):
|
38 |
+
if d.startswith("checkpoint-"):
|
39 |
+
dirs.append(d)
|
40 |
+
num_tot = len(dirs)
|
41 |
+
if num_tot <= num_keep:
|
42 |
+
return
|
43 |
+
# ensure ckpt is sorted and delete the ealier!
|
44 |
+
del_dirs = sorted(dirs, key=lambda x: int(x.split("-")[-1]))[: num_tot - num_keep]
|
45 |
+
for d in del_dirs:
|
46 |
+
path_to_dir = osp.join(base_path, d)
|
47 |
+
if osp.exists(path_to_dir):
|
48 |
+
shutil.rmtree(path_to_dir)
|
49 |
+
|
50 |
+
|
51 |
+
def save_videos_from_pil(pil_images, path, fps=8):
|
52 |
+
import av
|
53 |
+
|
54 |
+
save_fmt = Path(path).suffix
|
55 |
+
os.makedirs(os.path.dirname(path), exist_ok=True)
|
56 |
+
width, height = pil_images[0].size
|
57 |
+
|
58 |
+
if save_fmt == ".mp4":
|
59 |
+
codec = "libx264"
|
60 |
+
container = av.open(path, "w")
|
61 |
+
stream = container.add_stream(codec, rate=fps)
|
62 |
+
|
63 |
+
stream.width = width
|
64 |
+
stream.height = height
|
65 |
+
|
66 |
+
for pil_image in pil_images:
|
67 |
+
# pil_image = Image.fromarray(image_arr).convert("RGB")
|
68 |
+
av_frame = av.VideoFrame.from_image(pil_image)
|
69 |
+
container.mux(stream.encode(av_frame))
|
70 |
+
container.mux(stream.encode())
|
71 |
+
container.close()
|
72 |
+
|
73 |
+
elif save_fmt == ".gif":
|
74 |
+
pil_images[0].save(
|
75 |
+
fp=path,
|
76 |
+
format="GIF",
|
77 |
+
append_images=pil_images[1:],
|
78 |
+
save_all=True,
|
79 |
+
duration=(1 / fps * 1000),
|
80 |
+
loop=0,
|
81 |
+
)
|
82 |
+
else:
|
83 |
+
raise ValueError("Unsupported file type. Use .mp4 or .gif.")
|
84 |
+
|
85 |
+
|
86 |
+
def save_videos_grid(videos: torch.Tensor, path: str, rescale=False, n_rows=6, fps=8):
|
87 |
+
videos = rearrange(videos, "b c t h w -> t b c h w")
|
88 |
+
height, width = videos.shape[-2:]
|
89 |
+
outputs = []
|
90 |
+
|
91 |
+
for x in videos:
|
92 |
+
x = torchvision.utils.make_grid(x, nrow=n_rows) # (c h w)
|
93 |
+
x = x.transpose(0, 1).transpose(1, 2).squeeze(-1) # (h w c)
|
94 |
+
if rescale:
|
95 |
+
x = (x + 1.0) / 2.0 # -1,1 -> 0,1
|
96 |
+
x = (x * 255).numpy().astype(np.uint8)
|
97 |
+
x = Image.fromarray(x)
|
98 |
+
|
99 |
+
outputs.append(x)
|
100 |
+
|
101 |
+
os.makedirs(os.path.dirname(path), exist_ok=True)
|
102 |
+
|
103 |
+
save_videos_from_pil(outputs, path, fps)
|
104 |
+
|
105 |
+
|
106 |
+
def read_frames(video_path):
|
107 |
+
container = av.open(video_path)
|
108 |
+
|
109 |
+
video_stream = next(s for s in container.streams if s.type == "video")
|
110 |
+
frames = []
|
111 |
+
for packet in container.demux(video_stream):
|
112 |
+
for frame in packet.decode():
|
113 |
+
image = Image.frombytes(
|
114 |
+
"RGB",
|
115 |
+
(frame.width, frame.height),
|
116 |
+
frame.to_rgb().to_ndarray(),
|
117 |
+
)
|
118 |
+
frames.append(image)
|
119 |
+
|
120 |
+
return frames
|
121 |
+
|
122 |
+
|
123 |
+
def get_fps(video_path):
|
124 |
+
container = av.open(video_path)
|
125 |
+
video_stream = next(s for s in container.streams if s.type == "video")
|
126 |
+
fps = video_stream.average_rate
|
127 |
+
container.close()
|
128 |
+
return fps
|
src/vid2vid.py
ADDED
@@ -0,0 +1,233 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import os
|
3 |
+
import shutil
|
4 |
+
import ffmpeg
|
5 |
+
from datetime import datetime
|
6 |
+
from pathlib import Path
|
7 |
+
import numpy as np
|
8 |
+
import cv2
|
9 |
+
import torch
|
10 |
+
# import spaces
|
11 |
+
from diffusers import AutoencoderKL, DDIMScheduler
|
12 |
+
from einops import repeat
|
13 |
+
from omegaconf import OmegaConf
|
14 |
+
from PIL import Image
|
15 |
+
from torchvision import transforms
|
16 |
+
from transformers import CLIPVisionModelWithProjection
|
17 |
+
|
18 |
+
from src.models.pose_guider import PoseGuider
|
19 |
+
from src.models.unet_2d_condition import UNet2DConditionModel
|
20 |
+
from src.models.unet_3d import UNet3DConditionModel
|
21 |
+
from src.pipelines.pipeline_pose2vid_long import Pose2VideoPipeline
|
22 |
+
from src.utils.util import get_fps, read_frames, save_videos_grid
|
23 |
+
|
24 |
+
from src.utils.mp_utils import LMKExtractor
|
25 |
+
from src.utils.draw_util import FaceMeshVisualizer
|
26 |
+
from src.utils.pose_util import project_points_with_trans, matrix_to_euler_and_translation
|
27 |
+
from src.audio2vid import smooth_pose_seq
|
28 |
+
|
29 |
+
|
30 |
+
def parse_args():
|
31 |
+
parser = argparse.ArgumentParser()
|
32 |
+
parser.add_argument("--config", type=str, default='./configs/prompts/animation_facereenac.yaml')
|
33 |
+
parser.add_argument("-W", type=int, default=512)
|
34 |
+
parser.add_argument("-H", type=int, default=512)
|
35 |
+
parser.add_argument("-L", type=int)
|
36 |
+
parser.add_argument("--seed", type=int, default=42)
|
37 |
+
parser.add_argument("--cfg", type=float, default=3.5)
|
38 |
+
parser.add_argument("--steps", type=int, default=25)
|
39 |
+
parser.add_argument("--fps", type=int)
|
40 |
+
args = parser.parse_args()
|
41 |
+
|
42 |
+
return args
|
43 |
+
|
44 |
+
# @spaces.GPU
|
45 |
+
def video2video(ref_img, source_video, size=512, steps=25, length=150, seed=42):
|
46 |
+
cfg = 3.5
|
47 |
+
|
48 |
+
config = OmegaConf.load('./configs/prompts/animation_facereenac.yaml')
|
49 |
+
|
50 |
+
if config.weight_dtype == "fp16":
|
51 |
+
weight_dtype = torch.float16
|
52 |
+
else:
|
53 |
+
weight_dtype = torch.float32
|
54 |
+
|
55 |
+
vae = AutoencoderKL.from_pretrained(
|
56 |
+
config.pretrained_vae_path,
|
57 |
+
).to("cuda", dtype=weight_dtype)
|
58 |
+
|
59 |
+
reference_unet = UNet2DConditionModel.from_pretrained(
|
60 |
+
config.pretrained_base_model_path,
|
61 |
+
subfolder="unet",
|
62 |
+
).to(dtype=weight_dtype, device="cuda")
|
63 |
+
|
64 |
+
inference_config_path = config.inference_config
|
65 |
+
infer_config = OmegaConf.load(inference_config_path)
|
66 |
+
denoising_unet = UNet3DConditionModel.from_pretrained_2d(
|
67 |
+
config.pretrained_base_model_path,
|
68 |
+
config.motion_module_path,
|
69 |
+
subfolder="unet",
|
70 |
+
unet_additional_kwargs=infer_config.unet_additional_kwargs,
|
71 |
+
).to(dtype=weight_dtype, device="cuda")
|
72 |
+
|
73 |
+
pose_guider = PoseGuider(noise_latent_channels=320, use_ca=True).to(device="cuda", dtype=weight_dtype) # not use cross attention
|
74 |
+
|
75 |
+
image_enc = CLIPVisionModelWithProjection.from_pretrained(
|
76 |
+
config.image_encoder_path
|
77 |
+
).to(dtype=weight_dtype, device="cuda")
|
78 |
+
|
79 |
+
sched_kwargs = OmegaConf.to_container(infer_config.noise_scheduler_kwargs)
|
80 |
+
scheduler = DDIMScheduler(**sched_kwargs)
|
81 |
+
|
82 |
+
generator = torch.manual_seed(seed)
|
83 |
+
|
84 |
+
width, height = size, size
|
85 |
+
|
86 |
+
# load pretrained weights
|
87 |
+
denoising_unet.load_state_dict(
|
88 |
+
torch.load(config.denoising_unet_path, map_location="cpu"),
|
89 |
+
strict=False,
|
90 |
+
)
|
91 |
+
reference_unet.load_state_dict(
|
92 |
+
torch.load(config.reference_unet_path, map_location="cpu"),
|
93 |
+
)
|
94 |
+
pose_guider.load_state_dict(
|
95 |
+
torch.load(config.pose_guider_path, map_location="cpu"),
|
96 |
+
)
|
97 |
+
|
98 |
+
pipe = Pose2VideoPipeline(
|
99 |
+
vae=vae,
|
100 |
+
image_encoder=image_enc,
|
101 |
+
reference_unet=reference_unet,
|
102 |
+
denoising_unet=denoising_unet,
|
103 |
+
pose_guider=pose_guider,
|
104 |
+
scheduler=scheduler,
|
105 |
+
)
|
106 |
+
pipe = pipe.to("cuda", dtype=weight_dtype)
|
107 |
+
|
108 |
+
date_str = datetime.now().strftime("%Y%m%d")
|
109 |
+
time_str = datetime.now().strftime("%H%M")
|
110 |
+
save_dir_name = f"{time_str}--seed_{seed}-{size}x{size}"
|
111 |
+
|
112 |
+
save_dir = Path(f"output/{date_str}/{save_dir_name}")
|
113 |
+
save_dir.mkdir(exist_ok=True, parents=True)
|
114 |
+
|
115 |
+
|
116 |
+
lmk_extractor = LMKExtractor()
|
117 |
+
vis = FaceMeshVisualizer(forehead_edge=False)
|
118 |
+
|
119 |
+
|
120 |
+
|
121 |
+
ref_image_np = cv2.cvtColor(ref_img, cv2.COLOR_RGB2BGR)
|
122 |
+
# TODO: 人脸检测+裁剪
|
123 |
+
ref_image_np = cv2.resize(ref_image_np, (size, size))
|
124 |
+
ref_image_pil = Image.fromarray(cv2.cvtColor(ref_image_np, cv2.COLOR_BGR2RGB))
|
125 |
+
|
126 |
+
face_result = lmk_extractor(ref_image_np)
|
127 |
+
if face_result is None:
|
128 |
+
return None
|
129 |
+
|
130 |
+
lmks = face_result['lmks'].astype(np.float32)
|
131 |
+
ref_pose = vis.draw_landmarks((ref_image_np.shape[1], ref_image_np.shape[0]), lmks, normed=True)
|
132 |
+
|
133 |
+
|
134 |
+
|
135 |
+
source_images = read_frames(source_video)
|
136 |
+
src_fps = get_fps(source_video)
|
137 |
+
pose_transform = transforms.Compose(
|
138 |
+
[transforms.Resize((height, width)), transforms.ToTensor()]
|
139 |
+
)
|
140 |
+
|
141 |
+
step = 1
|
142 |
+
if src_fps == 60:
|
143 |
+
src_fps = 30
|
144 |
+
step = 2
|
145 |
+
|
146 |
+
pose_trans_list = []
|
147 |
+
verts_list = []
|
148 |
+
bs_list = []
|
149 |
+
src_tensor_list = []
|
150 |
+
args_L = len(source_images) if length==0 or length*step > len(source_images) else length*step
|
151 |
+
for src_image_pil in source_images[: args_L: step]:
|
152 |
+
src_tensor_list.append(pose_transform(src_image_pil))
|
153 |
+
src_img_np = cv2.cvtColor(np.array(src_image_pil), cv2.COLOR_RGB2BGR)
|
154 |
+
frame_height, frame_width, _ = src_img_np.shape
|
155 |
+
src_img_result = lmk_extractor(src_img_np)
|
156 |
+
if src_img_result is None:
|
157 |
+
break
|
158 |
+
pose_trans_list.append(src_img_result['trans_mat'])
|
159 |
+
verts_list.append(src_img_result['lmks3d'])
|
160 |
+
bs_list.append(src_img_result['bs'])
|
161 |
+
|
162 |
+
|
163 |
+
# pose_arr = np.array(pose_trans_list)
|
164 |
+
trans_mat_arr = np.array(pose_trans_list)
|
165 |
+
verts_arr = np.array(verts_list)
|
166 |
+
bs_arr = np.array(bs_list)
|
167 |
+
min_bs_idx = np.argmin(bs_arr.sum(1))
|
168 |
+
|
169 |
+
# compute delta pose
|
170 |
+
trans_mat_inv_frame_0 = np.linalg.inv(trans_mat_arr[0])
|
171 |
+
pose_arr = np.zeros([trans_mat_arr.shape[0], 6])
|
172 |
+
|
173 |
+
for i in range(pose_arr.shape[0]):
|
174 |
+
pose_mat = trans_mat_inv_frame_0 @ trans_mat_arr[i]
|
175 |
+
euler_angles, translation_vector = matrix_to_euler_and_translation(pose_mat)
|
176 |
+
pose_arr[i, :3] = euler_angles
|
177 |
+
pose_arr[i, 3:6] = translation_vector
|
178 |
+
|
179 |
+
pose_arr = smooth_pose_seq(pose_arr)
|
180 |
+
|
181 |
+
# face retarget
|
182 |
+
verts_arr = verts_arr - verts_arr[min_bs_idx] + face_result['lmks3d']
|
183 |
+
# project 3D mesh to 2D landmark
|
184 |
+
projected_vertices = project_points_with_trans(verts_arr, pose_arr, [frame_height, frame_width])
|
185 |
+
|
186 |
+
pose_list = []
|
187 |
+
for i, verts in enumerate(projected_vertices):
|
188 |
+
lmk_img = vis.draw_landmarks((frame_width, frame_height), verts, normed=False)
|
189 |
+
pose_image_np = cv2.resize(lmk_img, (width, height))
|
190 |
+
pose_list.append(pose_image_np)
|
191 |
+
|
192 |
+
pose_list = np.array(pose_list)
|
193 |
+
|
194 |
+
video_length = len(pose_list)
|
195 |
+
|
196 |
+
video = pipe(
|
197 |
+
ref_image_pil,
|
198 |
+
pose_list,
|
199 |
+
ref_pose,
|
200 |
+
width,
|
201 |
+
height,
|
202 |
+
video_length,
|
203 |
+
steps,
|
204 |
+
cfg,
|
205 |
+
generator=generator,
|
206 |
+
).videos
|
207 |
+
|
208 |
+
save_path = f"{save_dir}/{size}x{size}_{time_str}_noaudio.mp4"
|
209 |
+
save_videos_grid(
|
210 |
+
video,
|
211 |
+
save_path,
|
212 |
+
n_rows=1,
|
213 |
+
fps=src_fps,
|
214 |
+
)
|
215 |
+
|
216 |
+
audio_output = f'{save_dir}/audio_from_video.aac'
|
217 |
+
# extract audio
|
218 |
+
try:
|
219 |
+
ffmpeg.input(source_video).output(audio_output, acodec='copy').run()
|
220 |
+
# merge audio and video
|
221 |
+
stream = ffmpeg.input(save_path)
|
222 |
+
audio = ffmpeg.input(audio_output)
|
223 |
+
ffmpeg.output(stream.video, audio.audio, save_path.replace('_noaudio.mp4', '.mp4'), vcodec='copy', acodec='aac').run()
|
224 |
+
|
225 |
+
os.remove(save_path)
|
226 |
+
os.remove(audio_output)
|
227 |
+
except:
|
228 |
+
shutil.move(
|
229 |
+
save_path,
|
230 |
+
save_path.replace('_noaudio.mp4', '.mp4')
|
231 |
+
)
|
232 |
+
|
233 |
+
return save_path.replace('_noaudio.mp4', '.mp4')
|