-
Notifications
You must be signed in to change notification settings - Fork 60
/
Copy pathdatasets.py
284 lines (241 loc) · 10.5 KB
/
datasets.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
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
import os
from enum import Enum
from PIL import Image
import numpy as np
import torch
from torchvision import datasets, transforms
INTERP = 3
class Dataset(Enum):
C10 = 1
C100 = 2
STL10 = 3
IN128 = 4
PLACES205 = 5
def get_encoder_size(dataset):
if dataset in [Dataset.C10, Dataset.C100]:
return 32
if dataset == Dataset.STL10:
return 64
if dataset in [Dataset.IN128, Dataset.PLACES205]:
return 128
raise RuntimeError("Couldn't get encoder size, unknown dataset: {}".format(dataset))
def get_dataset(dataset_name):
try:
return Dataset[dataset_name.upper()]
except KeyError as e:
raise KeyError("Unknown dataset '" + dataset_name + "'. Must be one of "
+ ', '.join([d.name for d in Dataset]))
class RandomTranslateWithReflect:
'''
Translate image randomly
Translate vertically and horizontally by n pixels where
n is integer drawn uniformly independently for each axis
from [-max_translation, max_translation].
Fill the uncovered blank area with reflect padding.
'''
def __init__(self, max_translation):
self.max_translation = max_translation
def __call__(self, old_image):
xtranslation, ytranslation = np.random.randint(-self.max_translation,
self.max_translation + 1,
size=2)
xpad, ypad = abs(xtranslation), abs(ytranslation)
xsize, ysize = old_image.size
flipped_lr = old_image.transpose(Image.FLIP_LEFT_RIGHT)
flipped_tb = old_image.transpose(Image.FLIP_TOP_BOTTOM)
flipped_both = old_image.transpose(Image.ROTATE_180)
new_image = Image.new("RGB", (xsize + 2 * xpad, ysize + 2 * ypad))
new_image.paste(old_image, (xpad, ypad))
new_image.paste(flipped_lr, (xpad + xsize - 1, ypad))
new_image.paste(flipped_lr, (xpad - xsize + 1, ypad))
new_image.paste(flipped_tb, (xpad, ypad + ysize - 1))
new_image.paste(flipped_tb, (xpad, ypad - ysize + 1))
new_image.paste(flipped_both, (xpad - xsize + 1, ypad - ysize + 1))
new_image.paste(flipped_both, (xpad + xsize - 1, ypad - ysize + 1))
new_image.paste(flipped_both, (xpad - xsize + 1, ypad + ysize - 1))
new_image.paste(flipped_both, (xpad + xsize - 1, ypad + ysize - 1))
new_image = new_image.crop((xpad - xtranslation,
ypad - ytranslation,
xpad + xsize - xtranslation,
ypad + ysize - ytranslation))
return new_image
class TransformsC10:
'''
Apply the same input transform twice, with independent randomness.
'''
def __init__(self):
# flipping image along vertical axis
self.flip_lr = transforms.RandomHorizontalFlip(p=0.5)
# image augmentation functions
normalize = transforms.Normalize(mean=[x / 255.0 for x in [125.3, 123.0, 113.9]],
std=[x / 255.0 for x in [63.0, 62.1, 66.7]])
col_jitter = transforms.RandomApply([
transforms.ColorJitter(0.4, 0.4, 0.4, 0.2)], p=0.8)
img_jitter = transforms.RandomApply([
RandomTranslateWithReflect(4)], p=0.8)
rnd_gray = transforms.RandomGrayscale(p=0.25)
# main transform for self-supervised training
self.train_transform = transforms.Compose([
img_jitter,
col_jitter,
rnd_gray,
transforms.ToTensor(),
normalize
])
# transform for testing
self.test_transform = transforms.Compose([
transforms.ToTensor(),
normalize
])
def __call__(self, inp):
inp = self.flip_lr(inp)
out1 = self.train_transform(inp)
out2 = self.train_transform(inp)
return out1, out2
class TransformsSTL10:
'''
Apply the same input transform twice, with independent randomness.
'''
def __init__(self):
# flipping image along vertical axis
self.flip_lr = transforms.RandomHorizontalFlip(p=0.5)
normalize = transforms.Normalize(mean=(0.43, 0.42, 0.39), std=(0.27, 0.26, 0.27))
# image augmentation functions
col_jitter = transforms.RandomApply([
transforms.ColorJitter(0.4, 0.4, 0.4, 0.2)], p=0.8)
rnd_gray = transforms.RandomGrayscale(p=0.25)
rand_crop = \
transforms.RandomResizedCrop(64, scale=(0.3, 1.0), ratio=(0.7, 1.4),
interpolation=INTERP)
self.test_transform = transforms.Compose([
transforms.Resize(70, interpolation=INTERP),
transforms.CenterCrop(64),
transforms.ToTensor(),
normalize
])
self.train_transform = transforms.Compose([
rand_crop,
col_jitter,
rnd_gray,
transforms.ToTensor(),
normalize
])
def __call__(self, inp):
inp = self.flip_lr(inp)
out1 = self.train_transform(inp)
out2 = self.train_transform(inp)
return out1, out2
class TransformsImageNet128:
'''
ImageNet dataset, for use with 128x128 full image encoder.
'''
def __init__(self):
# image augmentation functions
self.flip_lr = transforms.RandomHorizontalFlip(p=0.5)
rand_crop = \
transforms.RandomResizedCrop(128, scale=(0.3, 1.0), ratio=(0.7, 1.4),
interpolation=INTERP)
col_jitter = transforms.RandomApply([
transforms.ColorJitter(0.4, 0.4, 0.4, 0.1)], p=0.8)
rnd_gray = transforms.RandomGrayscale(p=0.25)
post_transform = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
self.test_transform = transforms.Compose([
transforms.Resize(146, interpolation=INTERP),
transforms.CenterCrop(128),
post_transform
])
self.train_transform = transforms.Compose([
rand_crop,
col_jitter,
rnd_gray,
post_transform
])
def __call__(self, inp):
inp = self.flip_lr(inp)
out1 = self.train_transform(inp)
out2 = self.train_transform(inp)
return out1, out2
def build_dataset(dataset, batch_size, input_dir=None, labeled_only=False):
train_dir, val_dir = _get_directories(dataset, input_dir)
if dataset == Dataset.C10:
num_classes = 10
train_transform = TransformsC10()
test_transform = train_transform.test_transform
train_dataset = datasets.CIFAR10(root='/tmp/data/',
train=True,
transform=train_transform,
download=True)
test_dataset = datasets.CIFAR10(root='/tmp/data/',
train=False,
transform=test_transform,
download=True)
elif dataset == Dataset.C100:
num_classes = 100
train_transform = TransformsC10()
test_transform = train_transform.test_transform
train_dataset = datasets.CIFAR100(root='/tmp/data/',
train=True,
transform=train_transform,
download=True)
test_dataset = datasets.CIFAR100(root='/tmp/data/',
train=False,
transform=test_transform,
download=True)
elif dataset == Dataset.STL10:
num_classes = 10
train_transform = TransformsSTL10()
test_transform = train_transform.test_transform
train_split = 'train' if labeled_only else 'train+unlabeled'
train_dataset = datasets.STL10(root='/tmp/data/',
split=train_split,
transform=train_transform,
download=True)
test_dataset = datasets.STL10(root='/tmp/data/',
split='test',
transform=test_transform,
download=True)
elif dataset == Dataset.IN128:
num_classes = 1000
train_transform = TransformsImageNet128()
test_transform = train_transform.test_transform
train_dataset = datasets.ImageFolder(train_dir, train_transform)
test_dataset = datasets.ImageFolder(val_dir, test_transform)
elif dataset == Dataset.PLACES205:
num_classes = 1000
train_transform = TransformsImageNet128()
test_transform = train_transform.test_transform
train_dataset = datasets.ImageFolder(train_dir, train_transform)
test_dataset = datasets.ImageFolder(val_dir, test_transform)
# build pytorch dataloaders for the datasets
train_loader = \
torch.utils.data.DataLoader(dataset=train_dataset,
batch_size=batch_size,
shuffle=True,
pin_memory=True,
drop_last=True,
num_workers=16)
test_loader = \
torch.utils.data.DataLoader(dataset=test_dataset,
batch_size=batch_size,
shuffle=True,
pin_memory=True,
drop_last=True,
num_workers=16)
return train_loader, test_loader, num_classes
def _get_directories(dataset, input_dir):
if dataset in [Dataset.C10, Dataset.C100, Dataset.STL10]:
# Pytorch will download those datasets automatically
return None, None
if dataset == Dataset.IN128:
train_dir = os.path.join(input_dir, 'ILSVRC2012_img_train/')
val_dir = os.path.join(input_dir, 'ILSVRC2012_img_val/')
elif dataset == Dataset.PLACES205:
train_dir = os.path.join(input_dir, 'places205_256_train/')
val_dir = os.path.join(input_dir, 'places205_256_val/')
else:
raise 'Data directories for dataset ' + dataset + ' are not defined'
return train_dir, val_dir