File size: 9,256 Bytes
208b0eb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import os
import gc
import imageio
import numpy as np
import torch
import torchvision
import cv2
from einops import rearrange
from PIL import Image

def get_width_and_height_from_image_and_base_resolution(image, base_resolution):
    target_pixels = int(base_resolution) * int(base_resolution)
    original_width, original_height = Image.open(image).size
    ratio = (target_pixels / (original_width * original_height)) ** 0.5
    width_slider = round(original_width * ratio)
    height_slider = round(original_height * ratio)
    return height_slider, width_slider

def color_transfer(sc, dc):
    """
    Transfer color distribution from of sc, referred to dc.

    Args:
        sc (numpy.ndarray): input image to be transfered.
        dc (numpy.ndarray): reference image

    Returns:
        numpy.ndarray: Transferred color distribution on the sc.
    """

    def get_mean_and_std(img):
        x_mean, x_std = cv2.meanStdDev(img)
        x_mean = np.hstack(np.around(x_mean, 2))
        x_std = np.hstack(np.around(x_std, 2))
        return x_mean, x_std

    sc = cv2.cvtColor(sc, cv2.COLOR_RGB2LAB)
    s_mean, s_std = get_mean_and_std(sc)
    dc = cv2.cvtColor(dc, cv2.COLOR_RGB2LAB)
    t_mean, t_std = get_mean_and_std(dc)
    img_n = ((sc - s_mean) * (t_std / s_std)) + t_mean
    np.putmask(img_n, img_n > 255, 255)
    np.putmask(img_n, img_n < 0, 0)
    dst = cv2.cvtColor(cv2.convertScaleAbs(img_n), cv2.COLOR_LAB2RGB)
    return dst

def save_videos_grid(videos: torch.Tensor, path: str, rescale=False, n_rows=6, fps=12, imageio_backend=True, color_transfer_post_process=False):
    videos = rearrange(videos, "b c t h w -> t b c h w")
    outputs = []
    for x in videos:
        x = torchvision.utils.make_grid(x, nrow=n_rows)
        x = x.transpose(0, 1).transpose(1, 2).squeeze(-1)
        if rescale:
            x = (x + 1.0) / 2.0  # -1,1 -> 0,1
        x = (x * 255).numpy().astype(np.uint8)
        outputs.append(Image.fromarray(x))

    if color_transfer_post_process:
        for i in range(1, len(outputs)):
            outputs[i] = Image.fromarray(color_transfer(np.uint8(outputs[i]), np.uint8(outputs[0])))

    os.makedirs(os.path.dirname(path), exist_ok=True)
    if imageio_backend:
        if path.endswith("mp4"):
            imageio.mimsave(path, outputs, fps=fps)
        else:
            imageio.mimsave(path, outputs, duration=(1000 * 1/fps))
    else:
        if path.endswith("mp4"):
            path = path.replace('.mp4', '.gif')
        outputs[0].save(path, format='GIF', append_images=outputs, save_all=True, duration=100, loop=0)

def get_image_to_video_latent(validation_image_start, validation_image_end, video_length, sample_size):
    if validation_image_start is not None and validation_image_end is not None:
        if type(validation_image_start) is str and os.path.isfile(validation_image_start):
            image_start = clip_image = Image.open(validation_image_start).convert("RGB")
            image_start = image_start.resize([sample_size[1], sample_size[0]])
            clip_image = clip_image.resize([sample_size[1], sample_size[0]])
        else:
            image_start = clip_image = validation_image_start
            image_start = [_image_start.resize([sample_size[1], sample_size[0]]) for _image_start in image_start]
            clip_image = [_clip_image.resize([sample_size[1], sample_size[0]]) for _clip_image in clip_image]

        if type(validation_image_end) is str and os.path.isfile(validation_image_end):
            image_end = Image.open(validation_image_end).convert("RGB")
            image_end = image_end.resize([sample_size[1], sample_size[0]])
        else:
            image_end = validation_image_end
            image_end = [_image_end.resize([sample_size[1], sample_size[0]]) for _image_end in image_end]

        if type(image_start) is list:
            clip_image = clip_image[0]
            start_video = torch.cat(
                [torch.from_numpy(np.array(_image_start)).permute(2, 0, 1).unsqueeze(1).unsqueeze(0) for _image_start in image_start], 
                dim=2
            )
            input_video = torch.tile(start_video[:, :, :1], [1, 1, video_length, 1, 1])
            input_video[:, :, :len(image_start)] = start_video
            
            input_video_mask = torch.zeros_like(input_video[:, :1])
            input_video_mask[:, :, len(image_start):] = 255
        else:
            input_video = torch.tile(
                torch.from_numpy(np.array(image_start)).permute(2, 0, 1).unsqueeze(1).unsqueeze(0), 
                [1, 1, video_length, 1, 1]
            )
            input_video_mask = torch.zeros_like(input_video[:, :1])
            input_video_mask[:, :, 1:] = 255

        if type(image_end) is list:
            image_end = [_image_end.resize(image_start[0].size if type(image_start) is list else image_start.size) for _image_end in image_end]
            end_video = torch.cat(
                [torch.from_numpy(np.array(_image_end)).permute(2, 0, 1).unsqueeze(1).unsqueeze(0) for _image_end in image_end], 
                dim=2
            )
            input_video[:, :, -len(end_video):] = end_video
            
            input_video_mask[:, :, -len(image_end):] = 0
        else:
            image_end = image_end.resize(image_start[0].size if type(image_start) is list else image_start.size)
            input_video[:, :, -1:] = torch.from_numpy(np.array(image_end)).permute(2, 0, 1).unsqueeze(1).unsqueeze(0)
            input_video_mask[:, :, -1:] = 0

        input_video = input_video / 255

    elif validation_image_start is not None:
        if type(validation_image_start) is str and os.path.isfile(validation_image_start):
            image_start = clip_image = Image.open(validation_image_start).convert("RGB")
            image_start = image_start.resize([sample_size[1], sample_size[0]])
            clip_image = clip_image.resize([sample_size[1], sample_size[0]])
        else:
            image_start = clip_image = validation_image_start
            image_start = [_image_start.resize([sample_size[1], sample_size[0]]) for _image_start in image_start]
            clip_image = [_clip_image.resize([sample_size[1], sample_size[0]]) for _clip_image in clip_image]
        image_end = None
        
        if type(image_start) is list:
            clip_image = clip_image[0]
            start_video = torch.cat(
                [torch.from_numpy(np.array(_image_start)).permute(2, 0, 1).unsqueeze(1).unsqueeze(0) for _image_start in image_start], 
                dim=2
            )
            input_video = torch.tile(start_video[:, :, :1], [1, 1, video_length, 1, 1])
            input_video[:, :, :len(image_start)] = start_video
            input_video = input_video / 255
            
            input_video_mask = torch.zeros_like(input_video[:, :1])
            input_video_mask[:, :, len(image_start):] = 255
        else:
            input_video = torch.tile(
                torch.from_numpy(np.array(image_start)).permute(2, 0, 1).unsqueeze(1).unsqueeze(0), 
                [1, 1, video_length, 1, 1]
            ) / 255
            input_video_mask = torch.zeros_like(input_video[:, :1])
            input_video_mask[:, :, 1:, ] = 255
    else:
        image_start = None
        image_end = None
        input_video = torch.zeros([1, 3, video_length, sample_size[0], sample_size[1]])
        input_video_mask = torch.ones([1, 1, video_length, sample_size[0], sample_size[1]]) * 255
        clip_image = None

    del image_start
    del image_end
    gc.collect()

    return  input_video, input_video_mask, clip_image

def get_video_to_video_latent(input_video_path, video_length, sample_size, fps=None, validation_video_mask=None):
    if isinstance(input_video_path, str):
        cap = cv2.VideoCapture(input_video_path)
        input_video = []

        original_fps = cap.get(cv2.CAP_PROP_FPS)
        frame_skip = 1 if fps is None else int(original_fps // fps)

        frame_count = 0

        while True:
            ret, frame = cap.read()
            if not ret:
                break

            if frame_count % frame_skip == 0:
                frame = cv2.resize(frame, (sample_size[1], sample_size[0]))
                input_video.append(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))

            frame_count += 1

        cap.release()
    else:
        input_video = input_video_path

    input_video = torch.from_numpy(np.array(input_video))[:video_length]
    input_video = input_video.permute([3, 0, 1, 2]).unsqueeze(0) / 255

    if validation_video_mask is not None:
        validation_video_mask = Image.open(validation_video_mask).convert('L').resize((sample_size[1], sample_size[0]))
        input_video_mask = np.where(np.array(validation_video_mask) < 240, 0, 255)
        
        input_video_mask = torch.from_numpy(np.array(input_video_mask)).unsqueeze(0).unsqueeze(-1).permute([3, 0, 1, 2]).unsqueeze(0)
        input_video_mask = torch.tile(input_video_mask, [1, 1, input_video.size()[2], 1, 1])
        input_video_mask = input_video_mask.to(input_video.device, input_video.dtype)
    else:
        input_video_mask = torch.zeros_like(input_video[:, :1])
        input_video_mask[:, :, :] = 255

    return  input_video, input_video_mask, None