-
Notifications
You must be signed in to change notification settings - Fork 0
/
train_cifar.py
957 lines (792 loc) · 39.5 KB
/
train_cifar.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
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
import argparse
import logging
import sys
import time
import math
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
import os
from wideresnet_wavelet import WideResNetWavelet
from wideresnet import WideResNet
from preactresnet import PreActResNet18, PreActResNet50
from models import *
from utils import *
mu = torch.tensor(cifar10_mean).view(3,1,1).cuda()
std = torch.tensor(cifar10_std).view(3,1,1).cuda()
def normalize(X):
return (X - mu)/std
upper_limit, lower_limit = 1,0
def clamp(X, lower_limit, upper_limit):
return torch.max(torch.min(X, upper_limit), lower_limit)
class LabelSmoothingLoss(nn.Module):
def __init__(self, classes=10, smoothing=0.0, dim=-1):
super(LabelSmoothingLoss, self).__init__()
self.confidence = 1.0 - smoothing
self.smoothing = smoothing
self.cls = classes
self.dim = dim
def forward(self, pred, target):
pred = pred.log_softmax(dim=self.dim)
with torch.no_grad():
# true_dist = pred.data.clone()
true_dist = torch.zeros_like(pred)
true_dist.fill_(self.smoothing / (self.cls - 1))
true_dist.scatter_(1, target.data.unsqueeze(1), self.confidence)
return torch.mean(torch.sum(-true_dist * pred, dim=self.dim))
class Batches():
def __init__(self, dataset, batch_size, shuffle, set_random_choices=False, num_workers=0, drop_last=True):
self.dataset = dataset
self.batch_size = batch_size
self.set_random_choices = set_random_choices
self.dataloader = torch.utils.data.DataLoader(
dataset, batch_size=batch_size, num_workers=num_workers, pin_memory=True, shuffle=shuffle, drop_last=drop_last
)
def __iter__(self):
if self.set_random_choices:
self.dataset.set_random_choices()
return ({'input': x.to(device).float(), 'target': y.to(device).long()} for (x,y) in self.dataloader)
def __len__(self):
return len(self.dataloader)
def mixup_data(x, y, alpha=1.0):
'''Returns mixed inputs, pairs of targets, and lambda'''
if alpha > 0:
lam = np.random.beta(alpha, alpha)
else:
lam = 1
batch_size = x.size()[0]
index = torch.randperm(batch_size).cuda()
mixed_x = lam * x + (1 - lam) * x[index, :]
y_a, y_b = y, y[index]
return mixed_x, y_a, y_b, lam
def mixup_criterion(criterion, pred, y_a, y_b, lam):
return lam * criterion(pred, y_a) + (1 - lam) * criterion(pred, y_b)
def dlr_loss(x, y):
x_sorted, ind_sorted = x.sort(dim=1)
ind = (ind_sorted[:, -1] == y).float()
loss_value = -(x[np.arange(x.shape[0]), y] - x_sorted[:, -2] * ind - x_sorted[:, -1]
* (1. - ind)) / (x_sorted[:, -1] - x_sorted[:, -3] + 1e-12)
return loss_value.mean()
def CW_loss(x, y):
x_sorted, ind_sorted = x.sort(dim=1)
ind = (ind_sorted[:, -1] == y).float()
loss_value = -(x[np.arange(x.shape[0]), y] - x_sorted[:, -2] * ind - x_sorted[:, -1] * (1. - ind))
return loss_value.mean()
def attack_pgd(model, X, y, epsilon, alpha, attack_iters, restarts,
norm, mixup=False, y_a=None, y_b=None, lam=None,
early_stop=False, early_stop_pgd_max=1,
multitarget=False,
use_DLRloss=False, use_CWloss=False,
epoch=0, totalepoch=110, gamma=0.8,
use_adaptive=False, s_HE=15,
fast_better=False, BNeval=False):
max_loss = torch.zeros(y.shape[0]).cuda()
max_delta = torch.zeros_like(X).cuda()
if BNeval:
model.eval()
for _ in range(restarts):
# early stop pgd counter for each x
early_stop_pgd_count = early_stop_pgd_max * torch.ones(y.shape[0], dtype=torch.int32).cuda()
# initialize perturbation
delta = torch.zeros_like(X).cuda()
if norm == "l_inf":
delta.uniform_(-epsilon, epsilon)
elif norm == "l_2":
delta.normal_()
d_flat = delta.view(delta.size(0),-1)
n = d_flat.norm(p=2,dim=1).view(delta.size(0),1,1,1)
r = torch.zeros_like(n).uniform_(0, 1)
delta *= r/n*epsilon
else:
raise ValueError
delta = clamp(delta, lower_limit-X, upper_limit-X)
delta.requires_grad = True
iter_count = torch.zeros(y.shape[0])
# craft adversarial examples
for _ in range(attack_iters):
output = model(normalize(X + delta))
# if use early stop pgd
if early_stop:
# calculate mask for early stop pgd
if_success_fool = (output.max(1)[1] != y).to(dtype=torch.int32)
early_stop_pgd_count = early_stop_pgd_count - if_success_fool
index = torch.where(early_stop_pgd_count > 0)[0]
iter_count[index] = iter_count[index] + 1
else:
index = slice(None,None,None)
if not isinstance(index, slice) and len(index) == 0:
break
# Whether use mixup criterion
if fast_better:
loss_ori = F.cross_entropy(output, y)
grad_ori = torch.autograd.grad(loss_ori, delta, create_graph=True)[0]
loss_grad = (alpha / 4.) * (torch.norm(grad_ori.view(grad_ori.shape[0], -1), p=2, dim=1) ** 2)
loss = loss_ori + loss_grad.mean()
loss.backward()
grad = delta.grad.detach()
elif not mixup:
if multitarget:
random_label = torch.randint(low=0, high=10, size=y.shape).cuda()
random_direction = 2*((random_label == y).to(dtype=torch.float32) - 0.5)
loss = torch.mean(random_direction * F.cross_entropy(output, random_label, reduction='none'))
loss.backward()
grad = delta.grad.detach()
elif use_DLRloss:
beta_ = gamma * epoch / totalepoch
loss = (1. - beta_) * F.cross_entropy(output, y) + beta_ * dlr_loss(output, y)
loss.backward()
grad = delta.grad.detach()
elif use_CWloss:
beta_ = gamma * epoch / totalepoch
loss = (1. - beta_) * F.cross_entropy(output, y) + beta_ * CW_loss(output, y)
loss.backward()
grad = delta.grad.detach()
else:
if use_adaptive:
loss = F.cross_entropy(s_HE * output, y)
else:
#print(output.shape)
#print(y.shape)
loss = F.cross_entropy(output, y)
loss.backward()
grad = delta.grad.detach()
else:
criterion = nn.CrossEntropyLoss()
loss = mixup_criterion(criterion, model(normalize(X+delta)), y_a, y_b, lam)
loss.backward()
grad = delta.grad.detach()
d = delta[index, :, :, :]
g = grad[index, :, :, :]
x = X[index, :, :, :]
if norm == "l_inf":
d = torch.clamp(d + alpha * torch.sign(g), min=-epsilon, max=epsilon)
elif norm == "l_2":
g_norm = torch.norm(g.view(g.shape[0],-1),dim=1).view(-1,1,1,1)
scaled_g = g/(g_norm + 1e-10)
d = (d + scaled_g*alpha).view(d.size(0),-1).renorm(p=2,dim=0,maxnorm=epsilon).view_as(d)
d = clamp(d, lower_limit - x, upper_limit - x)
delta.data[index, :, :, :] = d
delta.grad.zero_()
if mixup:
criterion = nn.CrossEntropyLoss(reduction='none')
all_loss = mixup_criterion(criterion, model(normalize(X+delta)), y_a, y_b, lam)
else:
all_loss = F.cross_entropy(model(normalize(X+delta)), y, reduction='none')
max_delta[all_loss >= max_loss] = delta.detach()[all_loss >= max_loss]
max_loss = torch.max(max_loss, all_loss)
if BNeval:
model.train()
return max_delta, iter_count
def get_args():
parser = argparse.ArgumentParser()
parser.add_argument('--model', default='PreActResNet18')
parser.add_argument('--l1', default=0, type=float)
parser.add_argument('--data-dir', default='../cifar-data', type=str)
parser.add_argument('--epochs', default=110, type=int)
parser.add_argument('--lr-schedule', default='piecewise', choices=['superconverge', 'piecewise', 'linear', 'piecewisesmoothed', 'piecewisezoom', 'onedrop', 'multipledecay', 'cosine', 'cyclic'])
parser.add_argument('--lr-max', default=0.1, type=float)
parser.add_argument('--lr-one-drop', default=0.01, type=float)
parser.add_argument('--lr-drop-epoch', default=100, type=int)
parser.add_argument('--attack', default='pgd', type=str, choices=['pgd', 'fgsm', 'free', 'none'])
parser.add_argument('--epsilon', default=8, type=int)
parser.add_argument('--test_epsilon', default=8, type=int)
parser.add_argument('--attack-iters', default=10, type=int)
parser.add_argument('--restarts', default=1, type=int)
parser.add_argument('--pgd-alpha', default=2, type=float)
parser.add_argument('--test-pgd-alpha', default=2, type=float)
parser.add_argument('--fgsm-alpha', default=1.25, type=float)
parser.add_argument('--norm', default='l_inf', type=str, choices=['l_inf', 'l_2'])
parser.add_argument('--fgsm-init', default='random', choices=['zero', 'random', 'previous'])
parser.add_argument('--fname', default='cifar_model', type=str)
parser.add_argument('--seed', default=0, type=int)
parser.add_argument('--half', action='store_true')
parser.add_argument('--width-factor', default=10, type=int)
parser.add_argument('--resume', default=0, type=int)
parser.add_argument('--eval', action='store_true')
parser.add_argument('--val', action='store_true')
parser.add_argument('--chkpt-iters', default=100, type=int)
parser.add_argument('--mixture', action='store_true') # whether use mixture of clean and adv examples in a mini-batch
parser.add_argument('--mixture_alpha', type=float)
parser.add_argument('--l2', default=0, type=float)
# Group 1
parser.add_argument('--earlystopPGD', action='store_true') # whether use early stop in PGD
parser.add_argument('--earlystopPGDepoch1', default=60, type=int)
parser.add_argument('--earlystopPGDepoch2', default=100, type=int)
parser.add_argument('--warmup_lr', action='store_true') # whether warm_up lr from 0 to max_lr in the first n epochs
parser.add_argument('--warmup_lr_epoch', default=15, type=int)
parser.add_argument('--weight_decay', default=5e-4, type=float)#weight decay
parser.add_argument('--warmup_eps', action='store_true') # whether warm_up eps from 0 to 8/255 in the first n epochs
parser.add_argument('--warmup_eps_epoch', default=15, type=int)
parser.add_argument('--batch-size', default=128, type=int) #batch size
parser.add_argument('--labelsmooth', action='store_true') # whether use label smoothing
parser.add_argument('--labelsmoothvalue', default=0.0, type=float)
parser.add_argument('--lrdecay', default='base', type=str, choices=['intenselr', 'base', 'looselr', 'lineardecay'])
# Group 2
parser.add_argument('--use_DLRloss', action='store_true') # whether use DLRloss
parser.add_argument('--use_CWloss', action='store_true') # whether use CWloss
parser.add_argument('--use_multitarget', action='store_true') # whether use multitarget
parser.add_argument('--use_stronger_adv', action='store_true') # whether use mixture of clean and adv examples in a mini-batch
parser.add_argument('--stronger_index', default=0, type=int)
parser.add_argument('--use_FNandWN', action='store_true') # whether use FN and WN
parser.add_argument('--use_adaptive', action='store_true') # whether use s in attack during training
parser.add_argument('--s_FN', default=15, type=float) # s in FN
parser.add_argument('--m_FN', default=0.2, type=float) # s in FN
parser.add_argument('--use_FNonly', action='store_true') # whether use FN only
parser.add_argument('--fast_better', action='store_true')
parser.add_argument('--BNeval', action='store_true') # whether use eval mode for BN when crafting adversarial examples
parser.add_argument('--focalloss', action='store_true') # whether use focalloss
parser.add_argument('--focallosslambda', default=2., type=float)
parser.add_argument('--activation', default='ReLU', type=str)
parser.add_argument('--softplus_beta', default=1., type=float)
parser.add_argument('--optimizer', default='momentum', choices=['momentum', 'Nesterov', 'SGD_GC', 'SGD_GCC', 'Adam', 'AdamW'])
parser.add_argument('--mixup', action='store_true')
parser.add_argument('--mixup-alpha', type=float)
parser.add_argument('--cutout', action='store_true')
parser.add_argument('--cutout-len', type=int)
return parser.parse_args()
def get_auto_fname(args):
names = args.model + '_' + args.lr_schedule + '_eps' + str(args.epsilon) + '_bs' + str(args.batch_size) + '_maxlr' + str(args.lr_max)
# Group 1
if args.earlystopPGD:
names = names + '_earlystopPGD' + str(args.earlystopPGDepoch1) + str(args.earlystopPGDepoch2)
if args.warmup_lr:
names = names + '_warmuplr' + str(args.warmup_lr_epoch)
if args.warmup_eps:
names = names + '_warmupeps' + str(args.warmup_eps_epoch)
if args.weight_decay != 5e-4:
names = names + '_wd' + str(args.weight_decay)
if args.labelsmooth:
names = names + '_ls' + str(args.labelsmoothvalue)
# Group 2
if args.use_stronger_adv:
names = names + '_usestrongeradv#' + str(args.stronger_index)
if args.use_multitarget:
names = names + '_usemultitarget'
if args.use_DLRloss:
names = names + '_useDLRloss'
if args.use_CWloss:
names = names + '_useCWloss'
if args.use_FNandWN:
names = names + '_HE' + 's' + str(args.s_FN) + 'm' + str(args.m_FN)
if args.use_adaptive:
names = names + 'adaptive'
if args.use_FNonly:
names = names + '_FNonly'
if args.fast_better:
names = names + '_fastbetter'
if args.activation != 'ReLU':
names = names + '_' + args.activation
if args.activation == 'Softplus':
names = names + str(args.softplus_beta)
if args.lrdecay != 'base':
names = names + '_' + args.lrdecay
if args.BNeval:
names = names + '_BNeval'
if args.focalloss:
names = names + '_focalloss' + str(args.focallosslambda)
if args.optimizer != 'momentum':
names = names + '_' + args.optimizer
if args.mixup:
names = names + '_mixup' + str(args.mixup_alpha)
if args.cutout:
names = names + '_cutout' + str(args.cutout_len)
if args.attack != 'pgd':
names = names + '_' + args.attack
print('File name: ', names)
return names
def main():
args = get_args()
if args.fname == 'auto':
names = get_auto_fname(args)
args.fname = 'trained_models/' + names
else:
args.fname = 'trained_models/' + args.fname
if not os.path.exists(args.fname):
os.makedirs(args.fname)
logger = logging.getLogger(__name__)
logging.basicConfig(
format='[%(asctime)s] - %(message)s',
datefmt='%Y/%m/%d %H:%M:%S',
level=logging.DEBUG,
handlers=[
logging.FileHandler(os.path.join(args.fname, 'eval.log' if args.eval else 'output.log')),
logging.StreamHandler()
])
logger.info(args)
# Set seed
np.random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
# Prepare data
transforms = [Crop(32, 32), FlipLR()]
if args.cutout:
transforms.append(Cutout(args.cutout_len, args.cutout_len))
if args.val:
try:
dataset = torch.load("cifar10_validation_split.pth")
except:
print("Couldn't find a dataset with a validation split, did you run "
"generate_validation.py?")
return
val_set = list(zip(transpose(dataset['val']['data']/255.), dataset['val']['labels']))
val_batches = Batches(val_set, args.batch_size, shuffle=False, num_workers=4)
else:
dataset = cifar10(args.data_dir)
train_set = list(zip(transpose(pad(dataset['train']['data'], 4)/255.),
dataset['train']['labels']))
train_set_x = Transform(train_set, transforms)
train_batches = Batches(train_set_x, args.batch_size, shuffle=True, set_random_choices=True, num_workers=4)
test_set = list(zip(transpose(dataset['test']['data']/255.), dataset['test']['labels']))
test_batches = Batches(test_set, args.batch_size, shuffle=False, num_workers=4)
# Set perturbations
epsilon = (args.epsilon / 255.)
test_epsilon = (args.test_epsilon / 255.)
pgd_alpha = (args.pgd_alpha / 255.)
test_pgd_alpha = (args.test_pgd_alpha / 255.)
# Set models
if args.model == 'VGG':
model = VGG('VGG19')
elif args.model == 'ResNet18':
model = ResNet18()
elif args.model == 'GoogLeNet':
model = GoogLeNet()
elif args.model == 'DenseNet121':
model = DenseNet121()
elif args.model == 'DenseNet201':
model = DenseNet201()
elif args.model == 'ResNeXt29':
model = ResNeXt29_2x64d()
elif args.model == 'ResNeXt29L':
model = ResNeXt29_32x4d()
elif args.model == 'MobileNet':
model = MobileNet()
elif args.model == 'MobileNetV2':
model = MobileNetV2()
elif args.model == 'DPN26':
model = DPN26()
elif args.model == 'DPN92':
model = DPN92()
elif args.model == 'ShuffleNetG2':
model = ShuffleNetG2()
elif args.model == 'SENet18':
model = SENet18()
elif args.model == 'ShuffleNetV2':
model = ShuffleNetV2(1)
elif args.model == 'EfficientNetB0':
model = EfficientNetB0()
elif args.model == 'PNASNetA':
model = PNASNetA()
elif args.model == 'RegNetX':
model = RegNetX_200MF()
elif args.model == 'RegNetLX':
model = RegNetX_400MF()
elif args.model == 'PreActResNet50':
model = PreActResNet50()
elif args.model == 'PreActResNet18':
model = PreActResNet18(normalize_only_FN=args.use_FNonly, normalize=args.use_FNandWN, scale=args.s_FN,
activation=args.activation, softplus_beta=args.softplus_beta)
elif args.model == 'WideResNet':
model = WideResNet(34, 10, widen_factor=10, dropRate=0.0, normalize=args.use_FNandWN,
activation=args.activation, softplus_beta=args.softplus_beta)
elif args.model == 'WideResNet_20':
model = WideResNet(34, 10, widen_factor=20, dropRate=0.0, normalize=args.use_FNandWN,
activation=args.activation, softplus_beta=args.softplus_beta)
elif args.model == "WideResNetWavelet":
model = WideResNetWavelet(34, 10, widen_factor=10, dropRate=0.0, normalize=args.use_FNandWN, activation=args.activation, softplus_beta=args.softplus_beta)
else:
raise ValueError("Unknown model")
model = nn.DataParallel(model).cuda()
model.train()
# Set training hyperparameters
if args.l2:
decay, no_decay = [], []
for name,param in model.named_parameters():
if 'bn' not in name and 'bias' not in name:
decay.append(param)
else:
no_decay.append(param)
params = [{'params':decay, 'weight_decay':args.l2},
{'params':no_decay, 'weight_decay': 0 }]
else:
params = model.parameters()
if args.lr_schedule == 'cyclic':
opt = torch.optim.Adam(params, lr=args.lr_max, betas=(0.9, 0.999), eps=1e-08, weight_decay=args.weight_decay)
else:
if args.optimizer == 'momentum':
opt = torch.optim.SGD(params, lr=args.lr_max, momentum=0.9, weight_decay=args.weight_decay)
elif args.optimizer == 'Nesterov':
opt = torch.optim.SGD(params, lr=args.lr_max, momentum=0.9, weight_decay=args.weight_decay, nesterov=True)
elif args.optimizer == 'SGD_GC':
opt = SGD_GC(params, lr=args.lr_max, momentum=0.9, weight_decay=args.weight_decay)
elif args.optimizer == 'SGD_GCC':
opt = SGD_GCC(params, lr=args.lr_max, momentum=0.9, weight_decay=args.weight_decay)
elif args.optimizer == 'Adam':
opt = torch.optim.Adam(params, lr=args.lr_max, betas=(0.9, 0.999), eps=1e-08, weight_decay=args.weight_decay)
elif args.optimizer == 'AdamW':
opt = torch.optim.AdamW(params, lr=args.lr_max, betas=(0.9, 0.999), eps=1e-08, weight_decay=args.weight_decay)
# Cross-entropy (mean)
if args.labelsmooth:
criterion = LabelSmoothingLoss(smoothing=args.labelsmoothvalue)
else:
criterion = nn.CrossEntropyLoss()
# If we use freeAT or fastAT with previous init
if args.attack == 'free':
delta = torch.zeros(args.batch_size, 3, 32, 32).cuda()
delta.requires_grad = True
elif args.attack == 'fgsm' and args.fgsm_init == 'previous':
delta = torch.zeros(args.batch_size, 3, 32, 32).cuda()
delta.requires_grad = True
if args.attack == 'free':
epochs = int(math.ceil(args.epochs / args.attack_iters))
else:
epochs = args.epochs
# Set lr schedule
if args.lr_schedule == 'superconverge':
lr_schedule = lambda t: np.interp([t], [0, args.epochs * 2 // 5, args.epochs], [0, args.lr_max, 0])[0]
elif args.lr_schedule == 'piecewise':
def lr_schedule(t, warm_up_lr = args.warmup_lr):
if t < 100:
if warm_up_lr and t < args.warmup_lr_epoch:
return (t + 1.) / args.warmup_lr_epoch * args.lr_max
else:
return args.lr_max
if args.lrdecay == 'lineardecay':
if t < 105:
return args.lr_max * 0.02 * (105 - t)
else:
return 0.
elif args.lrdecay == 'intenselr':
if t < 102:
return args.lr_max / 10.
else:
return args.lr_max / 100.
elif args.lrdecay == 'looselr':
if t < 150:
return args.lr_max / 10.
else:
return args.lr_max / 100.
elif args.lrdecay == 'base':
if t < 105:
return args.lr_max / 10.
else:
return args.lr_max / 100.
elif args.lr_schedule == 'linear':
lr_schedule = lambda t: np.interp([t], [0, args.epochs // 3, args.epochs * 2 // 3, args.epochs], [args.lr_max, args.lr_max, args.lr_max / 10, args.lr_max / 100])[0]
elif args.lr_schedule == 'onedrop':
def lr_schedule(t):
if t < args.lr_drop_epoch:
return args.lr_max
else:
return args.lr_one_drop
elif args.lr_schedule == 'multipledecay':
def lr_schedule(t):
return args.lr_max - (t//(args.epochs//10))*(args.lr_max/10)
elif args.lr_schedule == 'cosine':
def lr_schedule(t):
return args.lr_max * 0.5 * (1 + np.cos(t / args.epochs * np.pi))
elif args.lr_schedule == 'cyclic':
def lr_schedule(t, stepsize=18, min_lr=1e-5, max_lr=args.lr_max):
# Scaler: we can adapt this if we do not want the triangular CLR
scaler = lambda x: 1.
# Additional function to see where on the cycle we are
cycle = math.floor(1 + t / (2 * stepsize))
x = abs(t / stepsize - 2 * cycle + 1)
relative = max(0, (1 - x)) * scaler(cycle)
return min_lr + (max_lr - min_lr) * relative
#### Set stronger adv attacks when decay the lr ####
def eps_alpha_schedule(t, warm_up_eps = args.warmup_eps, if_use_stronger_adv=args.use_stronger_adv, stronger_index=args.stronger_index): # Schedule number 0
if stronger_index == 0:
epsilon_s = [epsilon * 1.5, epsilon * 2]
pgd_alpha_s = [pgd_alpha, pgd_alpha]
elif stronger_index == 1:
epsilon_s = [epsilon * 1.5, epsilon * 2]
pgd_alpha_s = [pgd_alpha * 1.25, pgd_alpha * 1.5]
elif stronger_index == 2:
epsilon_s = [epsilon * 2, epsilon * 2.5]
pgd_alpha_s = [pgd_alpha * 1.5, pgd_alpha * 2]
else:
print('Undefined stronger index')
if if_use_stronger_adv:
if t < 100:
if t < args.warmup_eps_epoch and warm_up_eps:
return (t + 1.) / args.warmup_eps_epoch * epsilon, pgd_alpha, args.restarts
else:
return epsilon, pgd_alpha, args.restarts
elif t < 105:
return epsilon_s[0], pgd_alpha_s[0], args.restarts
else:
return epsilon_s[1], pgd_alpha_s[1], args.restarts
else:
if t < args.warmup_eps_epoch and warm_up_eps:
return (t + 1.) / args.warmup_eps_epoch * epsilon, pgd_alpha, args.restarts
else:
return epsilon, pgd_alpha, args.restarts
#### Set the counter for the early stop of PGD ####
def early_stop_counter_schedule(t):
if t < args.earlystopPGDepoch1:
return 1
elif t < args.earlystopPGDepoch2:
return 2
else:
return 3
best_test_robust_acc = 0
best_val_robust_acc = 0
if args.resume:
start_epoch = args.resume
model.load_state_dict(torch.load(os.path.join(args.fname, f'model_{start_epoch-1}.pth')))
opt.load_state_dict(torch.load(os.path.join(args.fname, f'opt_{start_epoch-1}.pth')))
logger.info(f'Resuming at epoch {start_epoch}')
best_test_robust_acc = torch.load(os.path.join(args.fname, f'model_best.pth'))['test_robust_acc']
if args.val:
best_val_robust_acc = torch.load(os.path.join(args.fname, f'model_val.pth'))['val_robust_acc']
else:
start_epoch = 0
if args.eval:
if not args.resume:
logger.info("No model loaded to evaluate, specify with --resume FNAME")
return
logger.info("[Evaluation mode]")
# logger.info('Epoch \t Train Time \t Test Time \t LR \t Train Loss \t Train Grad \t Train Acc \t Train Robust Loss \t Train Robust Acc || \t Test Loss \t Test Acc \t Test Robust Loss \t Test Robust Acc')
logger.info('Epoch \t Train Acc \t Train Robust Acc \t Test Acc \t Test Robust Acc')
# Records per epoch for savetxt
train_loss_record = []
train_acc_record = []
train_robust_loss_record = []
train_robust_acc_record = []
train_grad_record = []
test_loss_record = []
test_acc_record = []
test_robust_loss_record = []
test_robust_acc_record = []
test_grad_record = []
for epoch in range(start_epoch, epochs):
model.train()
start_time = time.time()
train_loss = 0
train_acc = 0
train_robust_loss = 0
train_robust_acc = 0
train_n = 0
train_grad = 0
record_iter = torch.tensor([])
for i, batch in enumerate(train_batches):
if args.eval:
break
X, y = batch['input'], batch['target']
onehot_target_withmargin_HE = args.m_FN * args.s_FN * torch.nn.functional.one_hot(y, num_classes=10)
if args.mixup:
X, y_a, y_b, lam = mixup_data(X, y, args.mixup_alpha)
X, y_a, y_b = map(Variable, (X, y_a, y_b))
epoch_now = epoch + (i + 1) / len(train_batches)
lr = lr_schedule(epoch_now)
opt.param_groups[0].update(lr=lr)
if args.attack == 'pgd':
# Random initialization
epsilon_sche, pgd_alpha_sche, restarts_sche = eps_alpha_schedule(epoch_now)
early_counter_max = early_stop_counter_schedule(epoch_now)
if args.mixup:
delta, iter_counts = attack_pgd(model, X, y, epsilon_sche, pgd_alpha_sche, args.attack_iters, restarts_sche, args.norm,
early_stop=args.earlystopPGD, early_stop_pgd_max=early_counter_max,
mixup=True, y_a=y_a, y_b=y_b, lam=lam)
else:
delta, iter_counts = attack_pgd(model, X, y, epsilon_sche, pgd_alpha_sche, args.attack_iters, restarts_sche, args.norm,
early_stop=args.earlystopPGD, early_stop_pgd_max=early_counter_max, multitarget=args.use_multitarget,
use_DLRloss=args.use_DLRloss, use_CWloss=args.use_CWloss,
epoch=epoch_now, totalepoch=args.epochs, gamma=0.8,
use_adaptive=args.use_adaptive, s_HE=args.s_FN,
fast_better=args.fast_better, BNeval=args.BNeval)
record_iter = torch.cat((record_iter, iter_counts))
delta = delta.detach()
elif args.attack == 'fgsm':
delta,_ = attack_pgd(model, X, y, epsilon, args.fgsm_alpha*epsilon, 1, 1, args.norm, fast_better=args.fast_better)
delta = delta.detach()
# Standard training
elif args.attack == 'none':
delta = torch.zeros_like(X)
adv_input = normalize(torch.clamp(X + delta[:X.size(0)], min=lower_limit, max=upper_limit))
adv_input.requires_grad = True
robust_output = model(adv_input)
# Training losses
if args.mixup:
clean_input = normalize(X)
clean_input.requires_grad = True
output = model(clean_input)
robust_loss = mixup_criterion(criterion, robust_output, y_a, y_b, lam)
elif args.mixture:
clean_input = normalize(X)
clean_input.requires_grad = True
output = model(clean_input)
robust_loss = args.mixture_alpha * criterion(robust_output, y) + (1-args.mixture_alpha) * criterion(output, y)
else:
clean_input = normalize(X)
clean_input.requires_grad = True
output = model(clean_input)
if args.focalloss:
criterion_nonreduct = nn.CrossEntropyLoss(reduction='none')
robust_confidence = F.softmax(robust_output, dim=1)[:, y].detach()
robust_loss = (criterion_nonreduct(robust_output, y) * ((1. - robust_confidence) ** args.focallosslambda)).mean()
elif args.use_DLRloss:
beta_ = 0.8 * epoch_now / args.epochs
robust_loss = (1. - beta_) * F.cross_entropy(robust_output, y) + beta_ * dlr_loss(robust_output, y)
elif args.use_CWloss:
beta_ = 0.8 * epoch_now / args.epochs
robust_loss = (1. - beta_) * F.cross_entropy(robust_output, y) + beta_ * CW_loss(robust_output, y)
elif args.use_FNandWN:
#print('use FN and WN with margin')
robust_loss = criterion(args.s_FN * robust_output - onehot_target_withmargin_HE, y)
else:
robust_loss = criterion(robust_output, y)
if args.l1:
for name,param in model.named_parameters():
if 'bn' not in name and 'bias' not in name:
robust_loss += args.l1*param.abs().sum()
opt.zero_grad()
robust_loss.backward()
opt.step()
clean_input = normalize(X)
clean_input.requires_grad = True
output = model(clean_input)
if args.mixup:
loss = mixup_criterion(criterion, output, y_a, y_b, lam)
else:
loss = criterion(output, y)
# Get the gradient norm values
input_grads = torch.autograd.grad(loss, clean_input, create_graph=False)[0]
# Record the statstic values
train_robust_loss += robust_loss.item() * y.size(0)
train_robust_acc += (robust_output.max(1)[1] == y).sum().item()
train_loss += loss.item() * y.size(0)
train_acc += (output.max(1)[1] == y).sum().item()
train_n += y.size(0)
train_grad += input_grads.abs().sum()
train_time = time.time()
if args.earlystopPGD:
print('Iter mean: ', record_iter.mean().item(), ' Iter std: ', record_iter.std().item())
#print('Learning rate: ', lr)
#print('Eps: ', epsilon_sche)
# Evaluate on test data
model.eval()
test_loss = 0
test_acc = 0
test_robust_loss = 0
test_robust_acc = 0
test_n = 0
test_grad = 0
for i, batch in enumerate(test_batches):
X, y = batch['input'], batch['target']
# Random initialization
if args.attack == 'none':
delta = torch.zeros_like(X)
else:
delta, _ = attack_pgd(model, X, y, test_epsilon, test_pgd_alpha, args.attack_iters, args.restarts, args.norm, early_stop=False)
delta = delta.detach()
adv_input = normalize(torch.clamp(X + delta[:X.size(0)], min=lower_limit, max=upper_limit))
adv_input.requires_grad = True
robust_output = model(adv_input)
robust_loss = criterion(robust_output, y)
clean_input = normalize(X)
clean_input.requires_grad = True
output = model(clean_input)
loss = criterion(output, y)
# Get the gradient norm values
input_grads = torch.autograd.grad(loss, clean_input, create_graph=False)[0]
test_robust_loss += robust_loss.item() * y.size(0)
test_robust_acc += (robust_output.max(1)[1] == y).sum().item()
test_loss += loss.item() * y.size(0)
test_acc += (output.max(1)[1] == y).sum().item()
test_n += y.size(0)
test_grad += input_grads.abs().sum()
test_time = time.time()
if args.val:
val_loss = 0
val_acc = 0
val_robust_loss = 0
val_robust_acc = 0
val_n = 0
for i, batch in enumerate(val_batches):
X, y = batch['input'], batch['target']
# Random initialization
if args.attack == 'none':
delta = torch.zeros_like(X)
elif args.attack == 'pgd':
delta, _ = attack_pgd(model, X, y, test_epsilon, pgd_alpha, args.attack_iters, args.restarts, args.norm, early_stop=False)
elif args.attack == 'fgsm':
delta,_ = attack_pgd(model, X, y, epsilon, args.fgsm_alpha*epsilon, 1, 1, args.norm, fast_better=args.fast_better)
delta = delta.detach()
robust_output = model(normalize(torch.clamp(X + delta[:X.size(0)], min=lower_limit, max=upper_limit)))
robust_loss = criterion(robust_output, y)
output = model(normalize(X))
loss = criterion(output, y)
val_robust_loss += robust_loss.item() * y.size(0)
val_robust_acc += (robust_output.max(1)[1] == y).sum().item()
val_loss += loss.item() * y.size(0)
val_acc += (output.max(1)[1] == y).sum().item()
val_n += y.size(0)
if not args.eval:
# logger.info('%d \t %.1f \t %.1f \t %.4f \t %.4f \t %.4f \t %.4f \t %.4f \t %.4f \t %.4f %.4f \t %.4f \t %.4f',
# epoch, train_time - start_time, test_time - train_time, lr,
# train_loss/train_n, train_grad/train_n, train_acc/train_n, train_robust_loss/train_n, train_robust_acc/train_n,
# test_loss/test_n, test_acc/test_n, test_robust_loss/test_n, test_robust_acc/test_n)
logger.info('%d \t %.4f \t %.4f \t %.4f \t %.4f',
epoch, train_acc/train_n, train_robust_acc/train_n, test_acc/test_n, test_robust_acc/test_n)
# Save results
train_loss_record.append(train_loss/train_n)
train_acc_record.append(train_acc/train_n)
train_robust_loss_record.append(train_robust_loss/train_n)
train_robust_acc_record.append(train_robust_acc/train_n)
train_grad_record.append(train_grad/train_n)
np.savetxt(args.fname+'/train_loss_record.txt', np.array(train_loss_record))
np.savetxt(args.fname+'/train_acc_record.txt', np.array(train_acc_record))
np.savetxt(args.fname+'/train_robust_loss_record.txt', np.array(train_robust_loss_record))
np.savetxt(args.fname+'/train_robust_acc_record.txt', np.array(train_robust_acc_record))
np.savetxt(args.fname+'/train_grad_record.txt', np.array(train_grad_record))
test_loss_record.append(test_loss/train_n)
test_acc_record.append(test_acc/train_n)
test_robust_loss_record.append(test_robust_loss/train_n)
test_robust_acc_record.append(test_robust_acc/train_n)
test_grad_record.append(test_grad/train_n)
np.savetxt(args.fname+'/test_loss_record.txt', np.array(test_loss_record))
np.savetxt(args.fname+'/test_acc_record.txt', np.array(test_acc_record))
np.savetxt(args.fname+'/test_robust_loss_record.txt', np.array(test_robust_loss_record))
np.savetxt(args.fname+'/test_robust_acc_record.txt', np.array(test_robust_acc_record))
np.savetxt(args.fname+'/test_grad_record.txt', np.array(test_grad_record))
if args.val:
logger.info('validation %.4f \t %.4f \t %.4f \t %.4f',
val_loss/val_n, val_acc/val_n, val_robust_loss/val_n, val_robust_acc/val_n)
if val_robust_acc/val_n > best_val_robust_acc:
torch.save({
'state_dict':model.state_dict(),
'test_robust_acc':test_robust_acc/test_n,
'test_robust_loss':test_robust_loss/test_n,
'test_loss':test_loss/test_n,
'test_acc':test_acc/test_n,
'val_robust_acc':val_robust_acc/val_n,
'val_robust_loss':val_robust_loss/val_n,
'val_loss':val_loss/val_n,
'val_acc':val_acc/val_n,
}, os.path.join(args.fname, f'model_val.pth'))
best_val_robust_acc = val_robust_acc/val_n
# save checkpoint
if epoch > 99 or (epoch+1) % args.chkpt_iters == 0 or epoch+1 == epochs:
torch.save(model.state_dict(), os.path.join(args.fname, f'model_{epoch}.pth'))
torch.save(opt.state_dict(), os.path.join(args.fname, f'opt_{epoch}.pth'))
# save best
if test_robust_acc/test_n > best_test_robust_acc:
torch.save({
'state_dict':model.state_dict(),
'test_robust_acc':test_robust_acc/test_n,
'test_robust_loss':test_robust_loss/test_n,
'test_loss':test_loss/test_n,
'test_acc':test_acc/test_n,
}, os.path.join(args.fname, f'model_best.pth'))
best_test_robust_acc = test_robust_acc/test_n
else:
logger.info('%d \t %.1f \t \t %.1f \t \t %.4f \t %.4f \t %.4f \t %.4f \t \t %.4f \t \t %.4f \t %.4f \t %.4f \t \t %.4f',
epoch, train_time - start_time, test_time - train_time, -1,
-1, -1, -1, -1,
test_loss/test_n, test_acc/test_n, test_robust_loss/test_n, test_robust_acc/test_n)
return
if __name__ == "__main__":
main()