Spaces:
Sleeping
Sleeping
File size: 13,653 Bytes
bc3753a |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 |
import torch, uuid
import os, sys, shutil, platform
# from src.facerender.pirender_animate import AnimateFromCoeff_PIRender
from src.utils.preprocess import CropAndExtract
from src.test_audio2coeff import Audio2Coeff
from src.facerender.animate import AnimateFromCoeff
from src.generate_batch import get_data
from src.generate_facerender_batch import get_facerender_data
from src.utils.init_path import init_path
# from pydub import AudioSegment
# def mp3_to_wav(mp3_filename,wav_filename,frame_rate):
# mp3_file = AudioSegment.from_file(file=mp3_filename)
# mp3_file.set_frame_rate(frame_rate).export(wav_filename,format="wav")
class SadTalker():
def __init__(self, checkpoint_path='checkpoints', config_path='src/config', lazy_load=False):
if torch.cuda.is_available():
device = "cuda"
elif platform.system() == 'Darwin': # macos
device = "mps"
else:
device = "cpu"
self.device = device
os.environ['TORCH_HOME']= checkpoint_path
self.checkpoint_path = checkpoint_path
self.config_path = config_path
self.sadtalker_paths = init_path(checkpoint_path, self.config_path, 256, False, 'crop')
self.animate_from_coeff = AnimateFromCoeff(self.sadtalker_paths, self.device)
self.audio_to_coeff = Audio2Coeff(self.sadtalker_paths, self.device)
def test(self,
pic_path,
crop_pic_path,
first_coeff_path,
crop_info,
source_image, driven_audio, preprocess='crop',
still_mode=False, use_enhancer=False, batch_size=1, size=256,
pose_style = 0,
facerender='facevid2vid',
exp_scale=1.0,
use_ref_video = False,
ref_video = None,
ref_info = None,
use_idle_mode = False,
length_of_audio = 0, use_blink=True, fps=20,
result_dir='./results/'):
# print(self.sadtalker_paths)
# self.preprocess_model = CropAndExtract(self.sadtalker_paths, self.device)
# if facerender == 'facevid2vid' and self.device != 'mps':
# self.animate_from_coeff = AnimateFromCoeff(self.sadtalker_paths, self.device)
# elif facerender == 'pirender' or self.device == 'mps':
# self.animate_from_coeff = AnimateFromCoeff_PIRender(self.sadtalker_paths, self.device)
# facerender = 'pirender'
# else:
# raise(RuntimeError('Unknown model: {}'.format(facerender)))
# time_tag = str(uuid.uuid4())
# save_dir = os.path.join(result_dir, time_tag)
# os.makedirs(save_dir, exist_ok=True)
save_dir = result_dir
os.makedirs(save_dir, exist_ok=True)
# input_dir = os.path.join(save_dir, 'input')
# os.makedirs(input_dir, exist_ok=True)
# print(source_image)
# pic_path = os.path.join(input_dir, os.path.basename(source_image))
# shutil.copy(source_image, input_dir)
# if driven_audio is not None and os.path.isfile(driven_audio):
# audio_path = os.path.join(input_dir, os.path.basename(driven_audio))
# #### mp3 to wav
# if '.mp3' in audio_path:
# mp3_to_wav(driven_audio, audio_path.replace('.mp3', '.wav'), 16000)
# audio_path = audio_path.replace('.mp3', '.wav')
# else:
# shutil.move(driven_audio, input_dir)
# elif use_idle_mode:
# audio_path = os.path.join(input_dir, 'idlemode_'+str(length_of_audio)+'.wav') ## generate audio from this new audio_path
# from pydub import AudioSegment
# one_sec_segment = AudioSegment.silent(duration=1000*length_of_audio) #duration in milliseconds
# one_sec_segment.export(audio_path, format="wav")
# else:
# print(use_ref_video, ref_info)
# assert use_ref_video == True and ref_info == 'all'
# if use_ref_video and ref_info == 'all': # full ref mode
# ref_video_videoname = os.path.basename(ref_video)
# audio_path = os.path.join(save_dir, ref_video_videoname+'.wav')
# print('new audiopath:',audio_path)
# # if ref_video contains audio, set the audio from ref_video.
# cmd = r"ffmpeg -y -hide_banner -loglevel error -i %s %s"%(ref_video, audio_path)
# os.system(cmd)
# os.makedirs(save_dir, exist_ok=True)
#crop image and extract 3dmm from image
# first_frame_dir = os.path.join(save_dir, 'first_frame_dir')
# os.makedirs(first_frame_dir, exist_ok=True)
# first_coeff_path, crop_pic_path, crop_info = self.preprocess_model.generate(pic_path, first_frame_dir, preprocess, True, size)
# if first_coeff_path is None:
# raise AttributeError("No face is detected")
# if use_ref_video:
# print('using ref video for genreation')
# ref_video_videoname = os.path.splitext(os.path.split(ref_video)[-1])[0]
# ref_video_frame_dir = os.path.join(save_dir, ref_video_videoname)
# os.makedirs(ref_video_frame_dir, exist_ok=True)
# print('3DMM Extraction for the reference video providing pose')
# ref_video_coeff_path, _, _ = self.preprocess_model.generate(ref_video, ref_video_frame_dir, preprocess, source_image_flag=False)
# else:
# ref_video_coeff_path = None
# if use_ref_video:
# if ref_info == 'pose':
# ref_pose_coeff_path = ref_video_coeff_path
# ref_eyeblink_coeff_path = None
# elif ref_info == 'blink':
# ref_pose_coeff_path = None
# ref_eyeblink_coeff_path = ref_video_coeff_path
# elif ref_info == 'pose+blink':
# ref_pose_coeff_path = ref_video_coeff_path
# ref_eyeblink_coeff_path = ref_video_coeff_path
# elif ref_info == 'all':
# ref_pose_coeff_path = None
# ref_eyeblink_coeff_path = None
# else:
# raise('error in refinfo')
# else:
# ref_pose_coeff_path = None
# ref_eyeblink_coeff_path = None
ref_pose_coeff_path = None
ref_eyeblink_coeff_path = None
audio_path = driven_audio
# fps = 25
#audio2ceoff
# if use_ref_video and ref_info == 'all':
# coeff_path = ref_video_coeff_path # self.audio_to_coeff.generate(batch, save_dir, pose_style, ref_pose_coeff_path)
# else:
batch = get_data(first_coeff_path, audio_path, self.device, ref_eyeblink_coeff_path=ref_eyeblink_coeff_path, still=still_mode, \
idlemode=use_idle_mode, length_of_audio=length_of_audio, use_blink=use_blink, fps = fps) # longer audio?
coeff = self.audio_to_coeff.generate(batch, save_dir, pose_style, ref_pose_coeff_path)
#coeff2video
data = get_facerender_data(coeff, crop_pic_path, first_coeff_path, audio_path, batch_size, still_mode=still_mode, \
preprocess=preprocess, size=size, expression_scale = exp_scale, facemodel=facerender)
return_path = self.animate_from_coeff.generate(data, save_dir, pic_path, crop_info, enhancer='gfpgan' if use_enhancer else None, preprocess=preprocess, img_size=size, fps = fps)
# video_name = data['video_name']
# print(f'The generated video is named {video_name} in {save_dir}')
# del self.preprocess_model
# del self.audio_to_coeff
# del self.animate_from_coeff
if torch.cuda.is_available():
torch.cuda.empty_cache()
torch.cuda.synchronize()
import gc; gc.collect()
return return_path
def test2(self, source_image, driven_audio, preprocess='crop',
still_mode=False, use_enhancer=False, batch_size=1, size=256,
pose_style = 0,
facerender='facevid2vid',
exp_scale=1.0,
use_ref_video = False,
ref_video = None,
ref_info = None,
use_idle_mode = False,
length_of_audio = 0, use_blink=True, fps = 20,
result_dir='./results/'):
os.makedirs(result_dir, exist_ok=True)
self.sadtalker_paths = init_path(self.checkpoint_path, self.config_path, size, False, preprocess)
print(self.sadtalker_paths)
self.audio_to_coeff = Audio2Coeff(self.sadtalker_paths, self.device)
self.preprocess_model = CropAndExtract(self.sadtalker_paths, self.device)
self.animate_from_coeff = AnimateFromCoeff(self.sadtalker_paths, self.device)
time_tag = str(uuid.uuid4())
save_dir = os.path.join(result_dir, time_tag)
os.makedirs(save_dir, exist_ok=True)
input_dir = os.path.join(save_dir, 'input')
os.makedirs(input_dir, exist_ok=True)
print(source_image)
pic_path = os.path.join(input_dir, os.path.basename(source_image))
shutil.copy(source_image, input_dir)
if driven_audio is not None and os.path.isfile(driven_audio):
audio_path = os.path.join(input_dir, os.path.basename(driven_audio))
shutil.copy(driven_audio, input_dir)
elif use_idle_mode:
audio_path = os.path.join(input_dir, 'idlemode_'+str(length_of_audio)+'.wav') ## generate audio from this new audio_path
from pydub import AudioSegment
one_sec_segment = AudioSegment.silent(duration=1000*length_of_audio) #duration in milliseconds
one_sec_segment.export(audio_path, format="wav")
else:
assert driven_audio is not None, "No audio is given"
print(use_ref_video, ref_info)
assert use_ref_video == True and ref_info == 'all'
if use_ref_video and ref_info == 'all': # full ref mode
ref_video_videoname = os.path.basename(ref_video)
audio_path = os.path.join(save_dir, ref_video_videoname+'.wav')
print('new audiopath:',audio_path)
# if ref_video contains audio, set the audio from ref_video.
cmd = r"ffmpeg -y -hide_banner -loglevel error -i %s %s"%(ref_video, audio_path)
os.system(cmd)
os.makedirs(save_dir, exist_ok=True)
#crop image and extract 3dmm from image
first_frame_dir = os.path.join(save_dir, 'first_frame_dir')
os.makedirs(first_frame_dir, exist_ok=True)
first_coeff_path, crop_pic_path, crop_info = self.preprocess_model.generate(pic_path, first_frame_dir, preprocess, True, size)
print(first_coeff_path, crop_info)
if first_coeff_path is None:
raise AttributeError("No face is detected")
if use_ref_video:
print('using ref video for genreation')
ref_video_videoname = os.path.splitext(os.path.split(ref_video)[-1])[0]
ref_video_frame_dir = os.path.join(save_dir, ref_video_videoname)
os.makedirs(ref_video_frame_dir, exist_ok=True)
print('3DMM Extraction for the reference video providing pose')
ref_video_coeff_path, _, _ = self.preprocess_model.generate(ref_video, ref_video_frame_dir, preprocess, source_image_flag=False)
else:
ref_video_coeff_path = None
if use_ref_video:
if ref_info == 'pose':
ref_pose_coeff_path = ref_video_coeff_path
ref_eyeblink_coeff_path = None
elif ref_info == 'blink':
ref_pose_coeff_path = None
ref_eyeblink_coeff_path = ref_video_coeff_path
elif ref_info == 'pose+blink':
ref_pose_coeff_path = ref_video_coeff_path
ref_eyeblink_coeff_path = ref_video_coeff_path
elif ref_info == 'all':
ref_pose_coeff_path = None
ref_eyeblink_coeff_path = None
else:
raise('error in refinfo')
else:
ref_pose_coeff_path = None
ref_eyeblink_coeff_path = None
#audio2ceoff
if use_ref_video and ref_info == 'all':
coeff_path = ref_video_coeff_path # self.audio_to_coeff.generate(batch, save_dir, pose_style, ref_pose_coeff_path)
else:
batch = get_data(first_coeff_path, audio_path, self.device, ref_eyeblink_coeff_path=ref_eyeblink_coeff_path, still=still_mode, \
idlemode=use_idle_mode, length_of_audio=length_of_audio, use_blink=use_blink, fps = fps) # longer audio?
coeff_path = self.audio_to_coeff.generate(batch, save_dir, pose_style, ref_pose_coeff_path)
#coeff2video
data = get_facerender_data(coeff_path, crop_pic_path, first_coeff_path, audio_path, batch_size, still_mode=still_mode, \
preprocess=preprocess, size=size, expression_scale = exp_scale, facemodel=facerender)
return_path = self.animate_from_coeff.generate(data, save_dir, pic_path, crop_info, enhancer='gfpgan' if use_enhancer else None, preprocess=preprocess, img_size=size, fps = fps)
# video_name = data['video_name']
print(f'The generated video is saved in {return_path}')
del self.preprocess_model
# del self.audio_to_coeff
# del self.animate_from_coeff
if torch.cuda.is_available():
torch.cuda.empty_cache()
torch.cuda.synchronize()
import gc; gc.collect()
return return_path |