-
Notifications
You must be signed in to change notification settings - Fork 67
/
dataset.py
248 lines (171 loc) · 7.6 KB
/
dataset.py
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
import pandas as pd
import random
from PIL import Image
from torch.utils.data import Dataset
import glob
import os
from augmentations import AugmentationTransform
from PIL import ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
class Vox256(Dataset):
def __init__(self, split, transform=None, augmentation=False):
if split == 'train':
self.ds_path = './datasets/vox/train'
elif split == 'test':
self.ds_path = './datasets/vox/test'
else:
raise NotImplementedError
self.videos = os.listdir(self.ds_path)
self.augmentation = augmentation
if self.augmentation:
self.aug = AugmentationTransform(False, True, True)
else:
self.aug = None
self.transform = transform
def __getitem__(self, idx):
video_path = os.path.join(self.ds_path, self.videos[idx])
frames_paths = sorted(glob.glob(video_path + '/*.png'))
nframes = len(frames_paths)
items = random.sample(list(range(nframes)), 2)
img_source = Image.open(frames_paths[items[0]]).convert('RGB')
img_target = Image.open(frames_paths[items[1]]).convert('RGB')
if self.augmentation:
img_source, img_target = self.aug(img_source, img_target)
if self.transform is not None:
img_source = self.transform(img_source)
img_target = self.transform(img_target)
return img_source, img_target
def __len__(self):
return len(self.videos)
class Vox256_vox2german(Dataset):
def __init__(self, transform=None):
self.source_root = './datasets/german/'
self.driving_root = './datasets/vox/test/'
self.anno = pd.read_csv('pairs_annotations/german_vox.csv')
self.source_imgs = os.listdir(self.source_root)
self.transform = transform
def __getitem__(self, idx):
source_name = str('%03d' % self.anno['source'][idx])
driving_name = self.anno['driving'][idx]
source_vid_path = self.source_root + source_name
driving_vid_path = self.driving_root + driving_name
source_frame_path = sorted(glob.glob(source_vid_path + '/*.png'))[0]
driving_frames_path = sorted(glob.glob(driving_vid_path + '/*.png'))[:100]
source_img = self.transform(Image.open(source_frame_path).convert('RGB'))
driving_vid = [self.transform(Image.open(p).convert('RGB')) for p in driving_frames_path]
return source_img, driving_vid, source_name, driving_name
def __len__(self):
return len(self.source_imgs)
class Vox256_eval(Dataset):
def __init__(self, transform=None):
self.ds_path = './datasets/vox/test/'
self.videos = os.listdir(self.ds_path)
self.transform = transform
def __getitem__(self, idx):
vid_name = self.videos[idx]
video_path = os.path.join(self.ds_path, vid_name)
frames_paths = sorted(glob.glob(video_path + '/*.png'))
vid_target = [self.transform(Image.open(p).convert('RGB')) for p in frames_paths]
return vid_name, vid_target
def __len__(self):
return len(self.videos)
class Vox256_cross(Dataset):
def __init__(self, transform=None):
self.ds_path = './datasets/vox/test/'
self.videos = os.listdir(self.ds_path)
self.anno = pd.read_csv('pairs_annotations/vox256.csv')
self.transform = transform
def __getitem__(self, idx):
source_name = self.anno['source'][idx]
driving_name = self.anno['driving'][idx]
source_vid_path = os.path.join(self.ds_path, source_name)
driving_vid_path = os.path.join(self.ds_path, driving_name)
source_frame_path = sorted(glob.glob(source_vid_path + '/*.png'))[0]
driving_frames_path = sorted(glob.glob(driving_vid_path + '/*.png'))[:100]
source_img = self.transform(Image.open(source_frame_path).convert('RGB'))
driving_vid = [self.transform(Image.open(p).convert('RGB')) for p in driving_frames_path]
return source_img, driving_vid, source_name, driving_name
def __len__(self):
return len(self.videos)
class Taichi(Dataset):
def __init__(self, split, transform=None, augmentation=False):
if split == 'train':
self.ds_path = './datasets/taichi/train/'
else:
self.ds_path = './datasets/taichi/test/'
self.videos = os.listdir(self.ds_path)
self.augmentation = augmentation
if self.augmentation:
self.aug = AugmentationTransform(True, True, True)
else:
self.aug = None
self.transform = transform
def __getitem__(self, idx):
video_path = self.ds_path + self.videos[idx]
frames_paths = sorted(glob.glob(video_path + '/*.png'))
nframes = len(frames_paths)
items = random.sample(list(range(nframes)), 2)
img_source = Image.open(frames_paths[items[0]]).convert('RGB')
img_target = Image.open(frames_paths[items[1]]).convert('RGB')
if self.augmentation:
img_source, img_target = self.aug(img_source, img_target)
if self.transform is not None:
img_source = self.transform(img_source)
img_target = self.transform(img_target)
return img_source, img_target
def __len__(self):
return len(self.videos)
class Taichi_eval(Dataset):
def __init__(self, transform=None):
self.ds_path = './datasets/taichi/test/'
self.videos = os.listdir(self.ds_path)
self.transform = transform
def __getitem__(self, idx):
vid_name = self.videos[idx]
video_path = os.path.join(self.ds_path, vid_name)
frames_paths = sorted(glob.glob(video_path + '/*.png'))
vid_target = [self.transform(Image.open(p).convert('RGB')) for p in frames_paths]
return vid_name, vid_target
def __len__(self):
return len(self.videos)
class TED(Dataset):
def __init__(self, split, transform=None, augmentation=False):
if split == 'train':
self.ds_path = './datasets/ted/train/'
else:
self.ds_path = './datasets/ted/test/'
self.videos = os.listdir(self.ds_path)
self.augmentation = augmentation
if self.augmentation:
self.aug = AugmentationTransform(False, True, True)
else:
self.aug = None
self.transform = transform
def __getitem__(self, idx):
video_path = os.path.join(self.ds_path, self.videos[idx])
frames_paths = sorted(glob.glob(video_path + '/*.png'))
nframes = len(frames_paths)
items = random.sample(list(range(nframes)), 2)
img_source = Image.open(frames_paths[items[0]]).convert('RGB')
img_target = Image.open(frames_paths[items[1]]).convert('RGB')
if self.augmentation:
img_source, img_target = self.aug(img_source, img_target)
if self.transform is not None:
img_source = self.transform(img_source)
img_target = self.transform(img_target)
return img_source, img_target
def __len__(self):
return len(self.videos)
class TED_eval(Dataset):
def __init__(self, transform=None):
self.ds_path = './datasets/ted/test/'
self.videos = os.listdir(self.ds_path)
self.transform = transform
def __getitem__(self, idx):
vid_name = self.videos[idx]
video_path = os.path.join(self.ds_path, vid_name)
frames_paths = sorted(glob.glob(video_path + '/*.png'))
vid_target = [self.transform(Image.open(p).convert('RGB')) for p in frames_paths]
return vid_name, vid_target
def __len__(self):
return len(self.videos)