-
Notifications
You must be signed in to change notification settings - Fork 3
/
datasets.py
369 lines (273 loc) · 11.9 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
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
import torch.utils.data as data
from PIL import Image
import numpy as np
import torchvision
from torchvision.datasets import CIFAR10, SVHN, ImageFolder, DatasetFolder, utils
import os
import os.path
import logging
import random
import torch
logging.basicConfig()
logger = logging.getLogger()
logger.setLevel(logging.INFO)
IMG_EXTENSIONS = ('.jpg', '.jpeg', '.png', '.ppm', '.bmp', '.pgm', '.tif', '.tiff', '.webp')
def mkdirs(dirpath):
try:
os.makedirs(dirpath)
except Exception as _:
pass
class TriggerHandler(object):
def __init__(self, trigger_path, trigger_size, trigger_label, img_width, img_height):
self.trigger_img = Image.open(trigger_path).convert('RGB')
self.trigger_size = trigger_size
self.trigger_img = self.trigger_img.resize((trigger_size, trigger_size))
self.trigger_label = trigger_label
self.img_width = img_width
self.img_height = img_height
def put_trigger(self, img):
img.paste(self.trigger_img, (self.img_width - self.trigger_size, self.img_height - self.trigger_size))
return img
class CIFAR10_truncated(data.Dataset):
def __init__(self, root, dataidxs=None, train=True, transform=None, target_transform=None, download=False):
self.root = root
self.dataidxs = dataidxs
self.train = train
self.transform = transform
self.target_transform = target_transform
self.download = download
self.data, self.target = self.__build_truncated_dataset__()
def __build_truncated_dataset__(self):
cifar_dataobj = CIFAR10(self.root, self.train, self.transform, self.target_transform, self.download)
if torchvision.__version__ == '0.2.1':
if self.train:
data, target = cifar_dataobj.train_data, np.array(cifar_dataobj.train_labels)
else:
data, target = cifar_dataobj.test_data, np.array(cifar_dataobj.test_labels)
else:
data = cifar_dataobj.data
target = np.array(cifar_dataobj.targets)
if self.dataidxs is not None:
data = data[self.dataidxs]
target = target[self.dataidxs]
return data, target
def truncate_channel(self, index):
for i in range(index.shape[0]):
gs_index = index[i]
self.data[gs_index, :, :, 1] = 0.0
self.data[gs_index, :, :, 2] = 0.0
def __getitem__(self, index):
"""
Args:
index (int): Index
Returns:
tuple: (image, target) where target is index of the target class.
"""
img, target = self.data[index], self.target[index]
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
target = self.target_transform(target)
return img, target
def __len__(self):
return len(self.data)
class CIFAR10Poison_truncated(data.Dataset):
def __init__(self, args, root, dataidxs=None, train=True, transform=None, target_transform=None, download=False):
self.root = root
self.dataidxs = dataidxs
self.train = train
self.transform = transform
self.target_transform = target_transform
self.download = download
self.width = 32
self.height =32
self.channels = 3
self.trigger_handler = TriggerHandler(args.trigger_path, args.trigger_size, args.trigger_label, self.width, self.height)
self.poisoning_rate = args.poisoning_rate if train else 1.0
self.data, self.target = self.__build_truncated_dataset__()
indices = range(len(self.target))
self.poi_indices = random.sample(indices, k=int(len(indices) * self.poisoning_rate))
def __build_truncated_dataset__(self):
cifar_dataobj = CIFAR10(self.root, self.train, self.transform, self.target_transform, self.download)
if torchvision.__version__ == '0.2.1':
if self.train:
data, target = cifar_dataobj.train_data, np.array(cifar_dataobj.train_labels)
else:
data, target = cifar_dataobj.test_data, np.array(cifar_dataobj.test_labels)
else:
data = cifar_dataobj.data
target = np.array(cifar_dataobj.targets)
if self.dataidxs is not None:
data = data[self.dataidxs]
target = target[self.dataidxs]
return data, target
def truncate_channel(self, index):
for i in range(index.shape[0]):
gs_index = index[i]
self.data[gs_index, :, :, 1] = 0.0
self.data[gs_index, :, :, 2] = 0.0
def __getitem__(self, index):
"""
Args:
index (int): Index
Returns:
tuple: (image, target) where target is index of the target class.
"""
img, target = self.data[index], self.target[index]
img = Image.fromarray(img)
if index in self.poi_indices:
target = self.trigger_handler.trigger_label
img = self.trigger_handler.put_trigger(img)
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
target = self.target_transform(target)
return img, target
def __len__(self):
return len(self.data)
class SVHN_truncated(data.Dataset):
def __init__(self, root, dataidxs=None, split="train", transform=None, target_transform=None, download=False):
self.root = root
self.dataidxs = dataidxs
self.split = split
self.transform = transform
self.target_transform = target_transform
self.download = download
self.data, self.target = self.__build_truncated_dataset__()
def __build_truncated_dataset__(self):
svhn_dataobj = SVHN(self.root, self.split, self.transform, self.target_transform, self.download)
data = svhn_dataobj.data
target = np.array(svhn_dataobj.labels)
if self.dataidxs is not None:
data = data[self.dataidxs]
target = target[self.dataidxs]
return data, target
def truncate_channel(self, index):
for i in range(index.shape[0]):
gs_index = index[i]
self.data[gs_index, :, :, 1] = 0.0
self.data[gs_index, :, :, 2] = 0.0
def __getitem__(self, index):
img, target = self.data[index], int(self.target[index])
img = Image.fromarray(np.transpose(img, (1, 2, 0)))
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
target = self.target_transform(target)
return img, target
def __len__(self):
return len(self.data)
class SVHNPoison_truncated(data.Dataset):
def __init__(self, args, root, dataidxs=None, split="train", transform=None, target_transform=None, download=False):
self.root = root
self.dataidxs = dataidxs
self.split = split
self.transform = transform
self.target_transform = target_transform
self.download = download
self.data, self.target = self.__build_truncated_dataset__()
if split == "train":
train = True
else:
train = False
self.width = 32
self.height =32
self.channels = 3
self.trigger_handler = TriggerHandler(args.trigger_path, args.trigger_size, args.trigger_label, self.width, self.height)
self.poisoning_rate = args.poisoning_rate if train else 1.0
self.data, self.target = self.__build_truncated_dataset__()
indices = range(len(self.target))
self.poi_indices = random.sample(indices, k=int(len(indices) * self.poisoning_rate))
def __build_truncated_dataset__(self):
svhn_dataobj = SVHN(self.root, self.split, self.transform, self.target_transform, self.download)
data = svhn_dataobj.data
target = np.array(svhn_dataobj.labels)
if self.dataidxs is not None:
data = data[self.dataidxs]
target = target[self.dataidxs]
return data, target
def truncate_channel(self, index):
for i in range(index.shape[0]):
gs_index = index[i]
self.data[gs_index, :, :, 1] = 0.0
self.data[gs_index, :, :, 2] = 0.0
def __getitem__(self, index):
img, target = self.data[index], int(self.target[index])
img = Image.fromarray(np.transpose(img, (1, 2, 0)))
if index in self.poi_indices:
target = self.trigger_handler.trigger_label
img = self.trigger_handler.put_trigger(img)
if self.transform is not None:
img = self.transform(img)
if self.target_transform is not None:
target = self.target_transform(target)
return img, target
def __len__(self):
return len(self.data)
class ImageFolder_custom(DatasetFolder):
def __init__(self, root, dataidxs=None, train=True, transform=None, target_transform=None):
self.root = root
self.dataidxs = dataidxs
self.train = train
self.transform = transform
self.target_transform = target_transform
imagefolder_obj = ImageFolder(self.root, self.transform, self.target_transform)
self.loader = imagefolder_obj.loader
if self.dataidxs is not None:
self.samples = np.array(imagefolder_obj.samples)[self.dataidxs]
else:
self.samples = np.array(imagefolder_obj.samples)
self.target = self.samples[:,1]
def __getitem__(self, index):
path = self.samples[index][0]
target = self.target[index]
target = int(target)
sample = self.loader(path)
if self.transform is not None:
sample = self.transform(sample)
if self.target_transform is not None:
target = self.target_transform(target)
return sample, target
def __len__(self):
if self.dataidxs is None:
return len(self.samples)
else:
return len(self.dataidxs)
class ImageFolderPoison_custom(DatasetFolder):
def __init__(self, args, root, dataidxs=None, train=True, transform=None, target_transform=None):
self.root = root
self.dataidxs = dataidxs
self.train = train
self.transform = transform
self.target_transform = target_transform
self.width = 64
self.height = 64
self.channels = 3
self.trigger_handler = TriggerHandler(args.trigger_path, args.trigger_size, args.trigger_label, self.width, self.height)
self.poisoning_rate = args.poisoning_rate if train else 1.0
imagefolder_obj = ImageFolder(self.root, self.transform, self.target_transform)
self.loader = imagefolder_obj.loader
if self.dataidxs is not None:
self.samples = np.array(imagefolder_obj.samples)[self.dataidxs]
else:
self.samples = np.array(imagefolder_obj.samples)
indices = range(len(self.samples))
self.poi_indices = random.sample(indices, k=int(len(indices) * self.poisoning_rate))
def __getitem__(self, index):
path = self.samples[index][0]
target = self.samples[index][1]
target = int(target)
sample = self.loader(path)
if index in self.poi_indices:
target = self.trigger_handler.trigger_label
sample = self.trigger_handler.put_trigger(sample)
if self.transform is not None:
sample = self.transform(sample)
if self.target_transform is not None:
target = self.target_transform(target)
return sample, target
def __len__(self):
if self.dataidxs is None:
return len(self.samples)
else:
return len(self.dataidxs)