-
Notifications
You must be signed in to change notification settings - Fork 2
/
Dataset.py
758 lines (694 loc) · 35 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
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
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
"""Dataset.py
The conversion of date sets, including 'CIFAR10', 'CIFAR100', 'MNIST', 'STL10', 'TINYIMAGENET'
The conversion methods include generating the validation set and reducing the numbers of training samples.
The labels for each data set are dense.
The benchmark training set is split into two subsets, including the training set (70% of the training set) and validation set (30% of the training set).
itapty_ily@hotmail.com
09.15.20
"""
import os
import pickle
import time
import numpy as np
import struct
import torch.utils.data as data
from PIL import Image as Image
# import PIL
import torch
import cv2
import torch.nn as nn
from Config import opt
from sklearn.utils import shuffle
import scipy.misc as misc
import torchvision as tv
import torchvision.transforms as transforms
import random
class Conversion:
"""conversion operation class.
Specify the date set and the number of training samples and save the converted date set in the new file.
Besides, we must re-write the codes in torchvision.dataset.XXXX to point at the path to modified data set files.
"""
def __init__(self, train_num=1000, dataset='mnist'):
print(self.__class__.__name__, 'is initializing')
dataset = dataset.upper()
if dataset in ['CIFAR10', 'CIFAR100', 'MNIST', 'STL10', 'TINYIMAGENET']:
self.dataset = dataset.upper()
else:
raise ValueError("Parameter dataset must be in 'CIFAR10', 'CIFAR100', 'MNIST', 'STL10'")
self.train_num = train_num
self._set_dataset_info()
def _set_dataset_info(self):
print(os.path.split(os.path.abspath('..')))
self._cifar10_dir = os.path.abspath('.') + '/Dataset/Cifar/cifar-10-batches-py/'
self._cifar100_dir = os.path.abspath('.') + '/Dataset/Cifar/cifar-100-python/'
self._mnist_dir = os.path.abspath('.') + '/Dataset/Mnist/'
self._stl10_dir = os.path.abspath('.') + '/Dataset/STL-10/'
self._tinyimagenet_dir = os.path.abspath('.') + '/Dataset/TinyImagenet/'
if self.dataset.upper() == 'CIFAR10':
print('The data set is cifar10')
self.img_size = [32, 32]
self.class_num = 10
self.img_channel = 3
self.num_test_set = 10000
if self.train_num > 50000:
self.train_num = 50000
self.num_train_set = int(self.train_num * 7 / 10)
self.num_valid_set = int(self.train_num * 3 / 10)
info = (self.dataset.upper(), self.num_train_set+self.num_valid_set, self.dataset.upper())
self.validset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_valid.pkl' % info
self.trainset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_tra.pkl' % info
self.testset = os.path.abspath('.') + '/Data_Files/%s/%s_tes.pkl' % self.dataset.upper()
if opt.embed == 'WEV':
file = os.path.join(self._cifar10_dir, 'batches.meta')
f = open(file, 'rb')
self.label_list = pickle.load(f)['label_names']
if self.dataset.upper() == 'CIFAR100':
self.img_size = [32, 32]
self.class_num = 100
self.img_channel = 3
self.num_test_set = 10000
if self.train_num > 50000:
self.train_num = 50000
self.num_train_set = int(self.train_num * 7 / 10)
self.num_valid_set = int(self.train_num * 3 / 10)
print('The data set is cifar100')
info = (self.dataset.upper(), self.num_train_set+self.num_valid_set, self.dataset.upper())
self.validset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_valid.pkl' % info
self.trainset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_tra.pkl' % info
self.testset = os.path.abspath('.') + '/Data_Files/%s/%s_tes.pkl' % self.dataset.upper()
if opt.embed == 'WEV':
file = os.path.join(self._cifar100_dir, 'meta')
f = open(file, 'rb')
self.label_list = pickle.load(f)['fine_label_names']
if self.dataset.upper() == 'MNIST':
print('The data set is mnist')
self.img_size = [28, 28]
self.class_num = 10
self.img_channel = 1
self.num_test_set = 10000
if self.train_num > 60000:
self.train_num = 60000
self.num_train_set = int(self.train_num * 7 / 10)
self.num_valid_set = int(self.train_num * 3 / 10)
info = (self.dataset.upper(), self.num_train_set+self.num_valid_set, self.dataset.upper(),)
self.validset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_valid.pkl' % info
self.trainset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_tra.pkl' % info
self.testset = os.path.abspath('.') + '/Data_Files/%s/%s_tes.pkl' % self.dataset.upper()
self.label_list = ['0', '1', '2', '3', '4',
'5', '6', '7', '8', '9']
if self.dataset.upper() == 'STL10':
print('The data set is STL10')
self.img_size = [96, 96]
self.class_num = 10
self.img_channel = 3
self. num_test_set = 8000
if self.train_num > 5000:
self.train_num = 5000
self.num_train_set = int(self.train_num * 7 / 10)
self.num_valid_set = int(self.train_num * 3 / 10)
info = (self.dataset.upper(), self.num_train_set+self.num_valid_set, self.dataset.upper())
self.validset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_valid.pkl' % info
self.trainset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_tra.pkl' % info
self.testset = os.path.abspath('.') + '/Data_Files/%s/%s_tes.pkl' % self.dataset.upper()
self.label_list = ['airplane', 'bird', 'car', 'cat', 'deer',
'dog', 'horse', 'monkey', 'ship', 'truck']
if self.dataset.upper() == 'TINYIMAGENET':
print('The data set is TINYIMAGENET')
self.img_size = [64, 64]
self.class_num = 200
self.train_num = self.train_num * self.class_num
self.img_channel = 3
self.num_test_set = 10000
# self.test_files = os.listdir(os.path.join(self._tinyimagenet_dir + 'test', 'images'))
if self.train_num > 100000:
self.train_num = 100000
self.num_train_set = self.train_num
self.num_valid_set = 10000
# print(self.train_num, self.num_train_set, self.num_valid_set)
info = (self.dataset.upper(), self.train_num, self.dataset.upper())
self.validset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_valid.pkl' % info
self.trainset = os.path.abspath('.') + '/Data_Files/%s/%d_%s_tra.pkl' % info
self.testset = os.path.abspath('.') + '/Data_Files/%s/%s_tes.pkl' % self.dataset.upper()
if opt.embed == 'WEV':
self.label_list = list()
self.wnids = open(os.path.join(self._tinyimagenet_dir, 'wnids.txt')).read().split('\n')
words = open(os.path.join(self._tinyimagenet_dir, 'words.txt')).read().split('\n')
for wnid in self.wnids:
for word in words:
if wnid in word:
self.label_list.append(word.split('\t')[1])
def _cv_cifar(self, dataset_path, files_list, ins_num=1000):
"""Convert a list of imgs in a py file to TFrecorder.
Convert a batch of imgs in a py file to TFrecorder.
Args:
dataset_path: the path to save recorder file.
files_list: the path to py files, save in a list.
"""
img_counter = list()
for i in range(self.class_num):
img_counter.append(0)
print('img_counter:', img_counter)
total_counter = 0
dataset_dict = dict()
dataset_dict['labels'] = []
dataset_dict['data'] = []
for f in files_list:
with open(f, 'rb')as file:
file_dict = pickle.load(file, encoding='latin1')
# print(list(file_dict.keys()))
imgs = file_dict['data']
labels = None
if self.dataset.upper() == 'CIFAR10':
labels = file_dict['labels']
elif self.dataset.upper() == 'CIFAR100':
labels = file_dict['fine_labels']
print('type(imgs):', type(imgs))
print('imgs.shape:', imgs.shape)
print('type(labels):', type(labels))
print('label.max VS label.min: ', np.max(labels), np.min(labels))
imgs_num = imgs.shape[0]
starttime = time.time()
for i in range(imgs_num):
img = imgs[i]
label = labels[i]
if i % 1000 == 0:
endtime = time.time()
print('Convert 1000 images cost %4.3fs.' % (endtime - starttime))
print('\tLoading...', os.path.split(f)[-1])
starttime = endtime
print('\t%4.2f percent finished.' % (sum(img_counter)/ins_num*100))
print('\timg_counter:', img_counter)
if img is not None:
flag_train = ('tra' in dataset_path) and (0 <= img_counter[int(label)] < int(ins_num / self.class_num))
flag_eval = ('valid' in dataset_path) and (self.num_train_set < total_counter + i) and (img_counter[int(label)] < int(ins_num / self.class_num))
if flag_train or ('tes' in dataset_path) or flag_eval:
img_counter[int(label)] += 1
dataset_dict['labels'].append(label)
dataset_dict['data'].append(img)
else:
pass
else:
raise ValueError('\t', i, '-th images is None')
total_counter += imgs_num
temp = os.path.join(os.path.join(os.path.abspath('.'), 'Data_files'), self.dataset)
if not os.path.exists(temp):
os.makedirs(temp)
dataset_dict['data'] = np.array(dataset_dict['data'])
dataset_dict['data'] = dataset_dict['data'].reshape((-1, 3, 32, 32))
dataset_dict['data'] = dataset_dict['data'].transpose((0, 2, 3, 1))
output = open(dataset_path, 'wb')
pickle.dump(dataset_dict, output)
output.close()
def _cv_mnist(self, dataset_path, files_list, ins_num=1000):
"""Convert a list of imgs in a py file to TFrecorder.
Convert a batch of imgs in a py file to TFrecorder.
Args:
dataset_path: the path to save recorder file.
files_list: the path to py files, save in a list.
"""
image_path, label_path = files_list
img_counter = list()
for i in range(self.class_num):
img_counter.append(0)
print('img_counter:', img_counter)
binfile = open(image_path, 'rb')
buffers = binfile.read()
head = struct.unpack_from('>IIII', buffers, 0)
print("head,", head)
offset = struct.calcsize('>IIII')
img_num = head[1]
width = head[2]
height = head[3]
bits = img_num * width * height
bits_string = '>' + str(bits) + 'B'
imgs = struct.unpack_from(bits_string, buffers, offset)
binfile.close()
imgs = np.reshape(imgs, [img_num, width, height])
print(np.max(imgs), np.min(imgs))
print('convert ', image_path, 'finished')
print('converting ', label_path)
binfile = open(label_path, 'rb')
buffers = binfile.read()
head = struct.unpack_from('>II', buffers, 0)
print("head,", head)
img_num = head[1]
offset = struct.calcsize('>II')
num_string = '>' + str(img_num) + "B"
labels = struct.unpack_from(num_string, buffers, offset)
binfile.close()
labels = np.reshape(labels, [img_num])
print('label.max VS label.min: ', np.max(labels), np.min(labels))
print('converting ', label_path, 'finished')
dataset_dict = dict()
dataset_dict['labels'] = []
dataset_dict['data'] = []
imgs_num = imgs.shape[0]
starttime = time.time()
for i in range(imgs_num):
img = imgs[i]
# print(img.shape)
label = labels[i]
if img is not None:
flag_train = ('tra' in dataset_path) and (0 <= img_counter[int(label)] < int(ins_num / self.class_num))
flag_eval = ('valid' in dataset_path) and (self.num_train_set < i) and (img_counter[int(label)] < int(ins_num / self.class_num))
if flag_train or ('tes' in dataset_path) or flag_eval:
img_counter[int(label)] += 1
dataset_dict['labels'].append(label)
dataset_dict['data'].append(img)
if i % 1000 == 0:
print('\n%0.2f-percent of images completed' % (sum(img_counter) / ins_num * 100))
endtime = time.time()
print('Convert 1000 images cost %4.3fs.' % (endtime - starttime))
starttime = endtime
print('img_counter:', img_counter)
# print('type(imgs):', type(imgs))
# print('imgs.shape:', imgs.shape)
# print('type(labels):', type(labels))
print('The label of the image is %f' % label)
show_img = np.reshape(np.array(img, dtype=np.uint8), (28, 28))
im = cv2.resize(show_img, (280, 280))
show_img = cv2.putText(im, str(label), (0, 40), cv2.FONT_HERSHEY_SIMPLEX, 1.2, (255, 255, 255), 2)
cv2.imshow('Resize', show_img)
k = cv2.waitKey(1) & 0xFF
if k == 27:
cv2.destroyAllWindows()
time.sleep(0.2)
else:
pass
else:
raise ValueError('\t', i, '-th images is None')
temp = os.path.join(os.path.join(os.path.abspath('.'), 'Data_files'), self.dataset)
if not os.path.exists(temp):
os.makedirs(temp)
dataset_dict['data'] = np.array(dataset_dict['data'])
dataset_dict['data'] = dataset_dict['data'].reshape((-1, 28, 28, 1))
# dataset_dict['data'] = dataset_dict['data'].transpose((0, 2, 3, 1))
output = open(dataset_path, 'wb')
pickle.dump(dataset_dict, output)
output.close()
cv2.destroyAllWindows()
def _cv_stl10(self, dataset_path, files_list, ins_num=1000):
"""Convert a list of imgs in a py file to TFrecorder.
Convert a batch of imgs in a py file to TFrecorder.
Args:
dataset_path: the path to save recorder file.
files_list: the path to py files, save in a list.
"""
image_path, label_path = files_list
img_counter = list()
for i in range(self.class_num):
img_counter.append(0)
print('img_counter:', img_counter)
images_file = open(image_path, 'rb')
labels_file = open(label_path, 'rb')
images_code = np.fromfile(images_file, dtype=np.uint8)
images = np.reshape(images_code, (-1, 3, 96, 96))
images = np.transpose(images, (0, 1, 3, 2))
labels = np.fromfile(labels_file, dtype=np.uint8)
print(images.shape)
print(labels.shape)
dataset_dict = dict()
dataset_dict['labels'] = []
dataset_dict['data'] = []
assert len(images) == len(labels)
print('label.max VS label.min: ', np.max(labels), np.min(labels))
starttime = time.time()
for i in range(len(images)):
img = images[i]
label = int(labels[i]-1)
if img is not None:
flag_train = ('tra' in dataset_path) and (0 <= img_counter[int(label)] < int(ins_num / self.class_num))
flag_eval = ('valid' in dataset_path) and (self.num_train_set < i) and (img_counter[int(label)] < int(ins_num / self.class_num))
if flag_train or ('tes' in dataset_path) or flag_eval:
img_counter[int(label)] += 1
dataset_dict['labels'].append(label)
dataset_dict['data'].append(img)
if i % 1000 == 0:
endtime = time.time()
print('Convert 1000 images cost %4.3fs.' % (endtime - starttime))
starttime = endtime
print('\t%4.2f percent finished.' % (sum(img_counter)/ins_num*100))
print('\timg_counter:', img_counter)
else:
pass
else:
raise ValueError('\t', i, '-th images is None')
temp = os.path.join(os.path.join(os.path.abspath('.'), 'Data_files'), self.dataset)
if not os.path.exists(temp):
os.makedirs(temp)
dataset_dict['data'] = np.array(dataset_dict['data'])
dataset_dict['data'] = dataset_dict['data'].reshape((-1, 3, 96, 96))
dataset_dict['data'] = dataset_dict['data'].transpose((0, 2, 3, 1))
output = open(dataset_path, 'wb')
pickle.dump(dataset_dict, output)
output.close()
def _cv_tinyimagenet(self, dataset_path, files_list, ins_num=1000):
"""Convert a list of imgs in a py file to TFrecorder.
Convert a batch of imgs in a py file to TFrecorder.
Args:
dataset_path: the path to save recorder file.
files_folder: the path to py files, save in a list.
"""
files_folder = files_list[0]
assert len(self.wnids) == self.class_num
dataset_dict = dict()
dataset_dict['labels'] = []
dataset_dict['data'] = []
if 'tra' in dataset_path:
img_counter = list()
for i in range(self.class_num):
img_counter.append(0)
print('img_counter:', img_counter)
for item in os.listdir(files_folder):
temp = os.path.join(os.path.join(files_folder, item), 'images')
image_files = os.listdir(temp)
random.seed(0)
random.shuffle(image_files)
if item in self.wnids:
label = self.wnids.index(item)
else:
raise ValueError('The label of %s is not in wnids.txt' % item)
for image_file in image_files:
image = cv2.imread(os.path.join(temp, image_file))
img_counter[label] += 1
dataset_dict['labels'].append(label)
dataset_dict['data'].append(image)
if img_counter[label] >= int(self.num_train_set / self.class_num):
break
print('%4.2f percent of images is converted.' % (sum(img_counter)/self.num_train_set*100))
print('img_counter:', img_counter)
dataset_dict['data'] = np.array(dataset_dict['data'])
elif 'valid' in dataset_path:
val_annotations = open(os.path.join(files_folder, 'val_annotations.txt')).read().split('\n')
i = 0
assert len(val_annotations) == self.num_valid_set
for val_annotation in val_annotations:
# print(val_annotation)
image_file, label_name = val_annotation.split('\t')[0:2]
temp = os.path.join(files_folder, 'images')
image = cv2.imread(os.path.join(temp, image_file))
label = self.wnids.index(label_name)
dataset_dict['labels'].append(label)
dataset_dict['data'].append(image)
i += 1
if i % 100 == 0:
print('%4.2f percent of images is converted.' % (i/self.num_valid_set*100))
# else:
# i = 0
# temp = os.path.join(files_folder, 'images')
# assert len(self.test_files) == self.num_test_set
# for image_file in self.test_files:
# image = cv2.imread(os.path.join(temp, image_file))
# dataset_dict['data'].append(image)
# dataset_dict['labels'].append(0)
# i += 1
# if i % 100 == 0:
# print('%4.2f percent of images is converted.' % (i/self.num_test_set*100))
temp = os.path.join(os.path.join(os.path.abspath('.'), 'Data_files'), self.dataset)
if not os.path.exists(temp):
os.makedirs(temp)
output = open(dataset_path, 'wb')
pickle.dump(dataset_dict, output)
output.close()
def dataset_to_pkl(self):
"""convert the cifar dataset bin files to TFRecoder with labels.
convert the cifar dataset bin files to TFRecoder with labels.
Return:
None
"""
if self.dataset.upper() == 'cifar10'.upper():
if not os.path.exists(self.trainset):
print('Converting training data')
self._cv_cifar(self.trainset,
files_list=[self._cifar10_dir+'data_batch_%d' % (i+1) for i in range(5)],
ins_num=self.num_train_set)
if not os.path.exists(self.validset):
print('Converting validation data')
self._cv_cifar(self.validset,
files_list=[self._cifar10_dir+'data_batch_%d' % (i+1) for i in range(5)],
ins_num=self.num_valid_set)
if not os.path.exists(self.testset):
print('Converting testing data')
self._cv_cifar(self.testset, files_list=[self._cifar10_dir + 'test_batch'])
elif self.dataset.upper() == 'cifar100'.upper():
if not os.path.exists(self.trainset):
print('Converting training data')
self._cv_cifar(self.trainset, files_list=[self._cifar100_dir + 'train'],
ins_num=self.num_train_set)
if not os.path.exists(self.validset):
print('Converting validation data')
self._cv_cifar(self.validset, files_list=[self._cifar100_dir + 'train'],
ins_num=self.num_valid_set)
if not os.path.exists(self.testset):
print('Converting testing data')
self._cv_cifar(self.testset, files_list=[self._cifar100_dir + 'test'])
elif self.dataset.upper() == 'mnist'.upper():
if not os.path.exists(self.trainset):
self._cv_mnist(self.trainset, [self._mnist_dir + 'train-images.idx3-ubyte',
self._mnist_dir + 'train-labels.idx1-ubyte'],
ins_num=self.num_train_set)
if not os.path.exists(self.validset):
self._cv_mnist(self.validset, [self._mnist_dir + 'train-images.idx3-ubyte',
self._mnist_dir + 'train-labels.idx1-ubyte'],
ins_num=self.num_valid_set)
if not os.path.exists(self.testset):
self._cv_mnist(self.testset, [self._mnist_dir + 't10k-images.idx3-ubyte',
self._mnist_dir + 't10k-labels.idx1-ubyte'])
elif self.dataset.upper() == 'stl10'.upper():
if not os.path.exists(self.trainset):
self._cv_stl10(self.trainset, [self._stl10_dir + 'train_X.bin',
self._stl10_dir + 'train_y.bin'],
ins_num=self.num_train_set)
if not os.path.exists(self.validset):
self._cv_stl10(self.validset, [self._stl10_dir + 'train_X.bin',
self._stl10_dir + 'train_y.bin'],
ins_num=self.num_valid_set)
if not os.path.exists(self.testset):
self._cv_stl10(self.testset, [self._stl10_dir + 'test_X.bin',
self._stl10_dir + 'test_y.bin'])
elif self.dataset.upper() == 'tinyimagenet'.upper():
if not os.path.exists(self.trainset):
self._cv_tinyimagenet(self.trainset, [
self._tinyimagenet_dir + 'train'],
ins_num=self.num_train_set)
if not os.path.exists(self.validset):
self._cv_tinyimagenet(self.validset, [
self._tinyimagenet_dir + 'val'],
ins_num=self.num_valid_set)
if not os.path.exists(self.testset):
self._cv_tinyimagenet(self.testset, [
self._tinyimagenet_dir + 'test'])
else:
raise ValueError('The input argues must be cifar10, cifar100, stl10, tinyimagenet or mnist')
def get_embeds(self, label_list=None):
dataset_folder = os.path.split(self.trainset)[0]
word_embed_file = os.path.join(dataset_folder, '%s_%d.pkl' % (opt.ResNet_blocks, self.train_num))
if not os.path.exists(word_embed_file):
file = open(self.trainset, 'rb')
temp = pickle.load(file)
file.close()
self.images = temp['data']
if 'ResNet' == opt.ResNet_blocks:
print("The prior LVs are sampled from ResNet")
if os.path.exists(os.path.join(os.path.abspath('.'), 'ResNetB_Model')) is False:
raise ValueError('The ResNetB model dose not exits. Please run ResNetB.py first.')
if opt.dataset.upper() == 'TINYIMAGENET'.upper():
from Resnet_imagenet import ResNet_blocks
z_x = ResNet_blocks(34, self.img_channel, self.img_size[0])
else:
from Resnet import ResNet_blocks
z_x = ResNet_blocks(5, self.img_channel, self.img_size[0])
z_x.load_state_dict(torch.load(os.path.join(os.path.join(os.path.abspath('.'),
'ResNetB_Model'),
'%s_%d_z_x' %
(opt.dataset.upper(), opt.train_num))))
z_x.eval()
if opt.ResNet_blocks == 'ResNet_AA':
print("The prior LVs are sampled from ResNet_AA")
if os.path.exists(os.path.join(os.path.abspath('.'), 'AA_Model')) is False:
raise ValueError('The AA model dose not exits. Please run AutoAugment.py first.')
if self.dataset.upper() in ['TINYIMAGENET'.upper()]:
from Resnet_imagenet import ResNet_blocks, ResNet_cf
z_x = ResNet_blocks(18, self.img_channel, self.img_size[0])
elif self.dataset.upper() in ['MNIST'.upper(), 'CIFAR10'.upper()]:
from Resnet import ResNet_blocks, ResNet_cf
z_x = ResNet_blocks(5, self.img_channel, self.img_size[0])
z_x.load_state_dict(torch.load(
os.path.join(os.path.join(os.path.abspath('.'), 'AA_Model'),\
'%s_%d_z_x' % (opt.dataset.upper(), opt.train_num))), strict=False)
z_x.eval()
self.word_embeds = list()
i = 0
z_x.cuda()
if 'AA' in opt.ResNet_blocks:
img_transformer = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465),
(0.2023, 0.1994, 0.2010)),
])
if self.dataset.upper() in ['MNIST'.upper()]:
img_transformer = transforms.Compose([
transforms.ToTensor(),
transforms.Normalize([0.4914],
[0.2023]),
])
else:
img_transformer = tv.transforms.ToTensor()
for image in self.images:
# print(image.dtype)
image = Image.fromarray(image.squeeze())
image = img_transformer(image).unsqueeze(0).cuda()
LV = z_x(image)
if type(LV) is tuple:
LV = LV[-1]
self.word_embeds.append(LV.detach().cpu())
i += 1
if i % 100 == 0:
print('%d LVs has been transformed' % i)
print('Totally %d LVs has been transformed' % i)
temp = os.path.join(os.path.join(os.path.abspath('.'), 'Data_files'), self.dataset)
if not os.path.exists(temp):
os.makedirs(temp)
output = open(word_embed_file, 'wb')
pickle.dump(self.word_embeds, output)
output.close()
else:
word_embed_file = open(word_embed_file, 'rb')
self.word_embeds = pickle.load(word_embed_file)
word_embed_file.close()
class Mydataset(data.Dataset, Conversion):
"""Definition of my modified dataset including the training, validation and test set.
Notes that the dataset is downloaded by user manually, and the dataset is saved in the following absolute path:
cifar10_dir = os.path.split(os.path.abspath('..'))[0] + '/Dataset/Cifar/cifar-10-batches-py/'
cifar100_dir = os.path.split(os.path.abspath('..'))[0] + '/Dataset/Cifar/cifar-100-python/'
mnist_dir = os.path.split(os.path.abspath('..'))[0] + '/Dataset/Mnist/'
stl10_dir = os.path.split(os.path.abspath('..'))[0] + '/Dataset/STL-10/'
Args:
train_num (int): the number of samples from training set, including training set and validation set:
training set = train_num % 70%
validation set = train_num % 30%
dataset (string): specify the data set to be used for our algorithm.
data_flag (string): specify which data set to be loaded,
including 'train', 'test' or 'valid'.
transform (callable, optional): A function/transform that takes in an PIL image
and returns a transformed version. E.g, ``transforms.RandomCrop``
target_transform (callable, optional): A function/transform that takes in the
target and transforms it.
"""
def __init__(self, train_num=1000, dataset='stl10', data_flag='train',
transform=None, target_transform=None, embed=False):
print(self.__class__.__name__, 'is initializing')
print(Conversion.__name__, 'is initializing')
Conversion.__init__(self, train_num=train_num, dataset=dataset)
self.transform = transform
self.target_transform = target_transform
self.train = data_flag.upper()
self.data_flag = data_flag
self.embed = embed
if self.data_flag == 'train':
self.data_file = self.trainset
elif self.data_flag == 'valid':
self.data_file = self.validset
elif self.data_flag == 'test':
self.data_file = self.testset
if not os.path.exists(self.data_file):
print('The data set files are not exist, we call dataset_to_pkl() to convert them.')
self.dataset_to_pkl()
if self.embed is True:
if opt.embed == 'WEV':
self.get_embeds(self.label_list)
else:
self.get_embeds()
file = open(self.data_file, 'rb')
temp = pickle.load(file)
self.images = temp['data']
self.labels = np.array(temp['labels'])
file.close()
if self.data_flag == 'train':
self.train_data = self.images
self.train_labels = self.labels
elif self.data_flag == 'valid':
self.valid_data = self.images
self.valid_labels = self.labels
elif self.data_flag == 'test':
self.test_data = self.images
self.test_labels = self.labels
"""
The images in the pkl files is correct, we can use the following codes to confirm it:
img = self.images[1000]
label = self.labels[1000]
show_img = np.reshape(np.array(img, dtype=np.uint8), (28, 28))
im = cv2.resize(show_img, (280, 280))
show_img = cv2.putText(im, str(label), (0, 40), cv2.FONT_HERSHEY_SIMPLEX, 1.2, (255, 255, 255), 2)
cv2.imshow('Resize', show_img)
k = cv2.waitKey(1) & 0xFF
if k == 27:
cv2.destroyAllWindows()
time.sleep(5)
"""
def __getitem__(self, index):
image = self.images[index]
label = self.labels[index]
"""
print(type(label)):
for MNIST: <class 'numpy.int32'> incorrect
"""
if self.dataset == 'MNIST':
# print(image.shape): (28, 28)
# print(np.max(image)): 255
image = np.array(image, dtype=np.byte).squeeze()
image = Image.fromarray(image, 'L')
label = int(label)
elif self.dataset == 'STL10':
# print(image.shape): (96, 96, 3)
# print(np.max(image)): 255
image = Image.fromarray(image)
label = int(label)
elif self.dataset == 'CIFAR10' or self.dataset == 'CIFAR100':
# print(image.shape): (32, 32, 3)
# print(np.max(image)): 255
image = Image.fromarray(image)
label = int(label)
elif self.dataset == 'TINYIMAGENET':
# print(image.shape): (64, 64, 3)
# print(np.max(image)): 256
image = Image.fromarray(image)
label = int(label)
if self.transform is not None:
for item in self.transform:
image = item(image)
if self.target_transform is not None:
label = self.target_transform(label)
"""
print(image.shape):
for MNIST: torch.Size([1, 28, 28]) correct
"""
if self.embed is True:
if opt.embed is not 'LV':
word_embed = self.word_embeds[label]
else:
word_embed = self.word_embeds[index]
return image, label, word_embed
else:
return image, label
def __len__(self):
if self.data_flag == 'train':
if len(self.images) == len(self.labels) == self.num_train_set:
return self.num_train_set
else:
print('len(self.images): ', len(self.images))
print('len(self.labels): ', len(self.labels))
print('self.num_train_set:: ', self.num_train_set)
raise ValueError('The length of data and label is incorrect.')
elif self.data_flag == 'valid':
if len(self.images) == len(self.labels) == self.num_valid_set:
return self.num_valid_set
else:
raise ValueError('The length of data and label is incorrect.')
elif self.data_flag == 'test':
if len(self.images) == len(self.labels) == self.num_test_set:
return self.num_test_set
else:
raise ValueError('The length of data and label is incorrect.')
if __name__ == '__main__':
pass