-
Notifications
You must be signed in to change notification settings - Fork 1
/
main.py
1354 lines (1046 loc) · 52.6 KB
/
main.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
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# import argparse
import os
import sys
import time
import shutil
import torch
import torch.nn.parallel
import torch.nn.functional as F
import torch.backends.cudnn as cudnn
import torch.optim
from torch.nn.utils import clip_grad_norm_
from utils.visualization import visualize_TSNE
from scipy.stats import genextreme as gev
from dataset import TSNDataSet, VideoRecord
from models import VideoModel
from loss import *
from opts import parser
from utils.utils import randSelectBatch
import math
from colorama import init
from colorama import Fore, Back, Style
import numpy as np
from tensorboardX import SummaryWriter
# auto hyperparameters tuning
import optuna
from optuna.trial import TrialState
from optuna.samplers import RandomSampler
import logging
os.environ["CUDA_VISIBLE_DEVICES"] = "0,1"
torch.backends.cudnn.enabled=False
# torch.cuda.set_device('1')
best_prec1 = 0
gpu_count = torch.cuda.device_count()
gpu_count = 1
def main(trial):
# To run CEVT model only, and get the similar result in paper,
# You only need to change code BLOCK 1 for dataset selecting,
# and BLOCK 2 for hyper-parameters selecting.
# Please search BLOCK 1 or BLOCK 2 in this file.
np.random.seed(1)
torch.manual_seed(1)
torch.cuda.manual_seed_all(1)
init(autoreset=True)
global args, best_prec1, writer
best_prec1 = 0
sys.argv += 'data/classInd_ucf_olympic.txt RGB /media/bigdata/uqyluo/MM2020Data/olympic/list_olympic_train_ucf_olympic-feature.txt /media/bigdata/uqyluo/MM2020Data/ucf101/list_ucf101_train_ucf_olympic-feature.txt /media/bigdata/uqyluo/MM2020Data/ucf101/list_ucf101_val_ucf_olympic-feature.txt --exp_path action/Testexp-SGD-share_params_Y-lr_3e-2-bS_32_129/ucf_olympic-16seg-disDA_none-alpha_0-advDA_none-beta_1_0.75_0.5-useBN_none-addlossDA_none-gamma_0.3-ensDA_none-mu_0-useAttn_none-n_attn_1/ --arch resnet101 --pretrained none --baseline_type video --frame_aggregation avgpool --num_segments 16 --val_segments 16 --add_fc 1 --fc_dim 512 --dropout_i 0.5 --dropout_v 0.5 --use_target uSv --share_params Y --dis_DA none --alpha 0 --place_dis N Y N --adv_DA none --beta 1 0.75 0.5 --place_adv N Y Y --use_bn none --add_loss_DA none --gamma 0.3 --ens_DA none --mu 0 --use_attn none --n_attn 1 --use_attn_frame none --gd 20 --lr 3e-2 --lr_decay 10 --lr_adaptive dann --lr_steps 10 20 --epochs 50 --optimizer SGD --n_rnn 1 --rnn_cell LSTM --n_directions 1 --n_ts 5 -b 32 129 32 -j 4 -ef 1 -pf 50 -sf 50 --copy_list N N'.split()
# arguments setting
parser.add_argument('--open_method', default="cls EVT", choices=['OSVM', 'cls OSVM', 'EVT', 'cls EVT'])
parser.add_argument('--OSVM_threshold', default=0.995) #args.OSVM_threshold python main.py
parser.add_argument('--EVT_threshold', default=0.53) # python main.py
parser.add_argument('--EVT_power', default=1)
parser.add_argument('--mix_interval_length', default=0.5)
# Entropy Maximization Loss
parser.add_argument('--entropy_max_loss', default=False)
parser.add_argument('--lambda_', default=1.5)
parser.add_argument('--adv_param', default=0.1)
parser.add_argument('--tsne', default=False)
parser.add_argument('--auto', default=False)
args = parser.parse_args()
########## BLOCK 1: Change Here for Different Datasets ##########
args.class_file = "data/classInd_ucf_olympic.txt"
args.train_source_list = "dataset/olympic/list_olympic_train_ucf_olympic-feature.txt"
args.train_target_list = "dataset/ucf101/list_ucf101_train_ucf_olympic-feature.txt"
args.val_list = "dataset/ucf101/list_ucf101_val_ucf_olympic-feature.txt"
########## END OF BLOCK 1 ##########
method = "CEVT" # CEVT | TA2N | TA3N | JAN | DAN | MCD | AdaBN
args.auto = False # True | False
if method == "CEVT":
args.entropy_max_loss = True # True | False
args.open_method = "cls EVT"
args.adv_DA = "RevGrad" # RevGrad | none
if args.auto:
args.lambda_ = trial.suggest_loguniform("lambda_", 1e-2, 1e1)
args.adv_param = trial.suggest_loguniform("adv_param", 1e-2, 2e1)
args.EVT_threshold = trial.suggest_uniform("EVT_threshold", 0.2, 0.7)
else:
########## BLOCK 2: Change Here for Different Hyper-Parameters ##########
args.lambda_ = 0.214 # H->U 0.7 | U->O 0.19 | O->U 0.214
args.adv_param = 5 # H->U 0.10 | U->O 1.83 | O->U 5
args.EVT_threshold = 0.3 # H->U 0.45 | U->O 0.565 | O->U 0.3
########## END OF BLOCK 2 ##########
elif method == "TA2N" or method == "TA3N":
args.frame_aggregation = 'trn-m'
args.adv_DA = "RevGrad"
args.use_attn = "TransAttn"
args.place_adv[0] = 'Y'
if args.auto:
if method == "TA3N":
args.add_loss_DA = "attentive_entropy"
args.gamma = trial.suggest_loguniform("c", 1e-3, 1e1)
segments = trial.suggest_int("segments", 2, 5)
args.OSVM_threshold = trial.suggest_uniform("OSVM_threshold", 0.6, 0.99)
else:
if method == "TA3N":
args.add_loss_DA = "attentive_entropy"
args.gamma = 0.009903869666099867
segments = 4
args.OSVM_threshold = 0.8990617407877891
args.num_segments = segments
args.val_segments = segments
elif method == "JAN" or method == "DAN":
args.dis_DA = method
args.alpha = trial.suggest_uniform("alpha", 0.01, 2) if args.auto else 0.1 # 1 | 0.1 | 0.01
args.OSVM_threshold = trial.suggest_uniform("OSVM_threshold", 0.5, 0.99) if args.auto else 0.99
elif method == "MCD" or method == "AdaBN":
args.ens_DA = method if method == "MCD" else "none"
args.use_bn = method if method == "AdaBN" else "none"
args.OSVM_threshold = trial.suggest_uniform("OSVM_threshold", 0.5, 0.99) if args.auto else 0.9999
if args.auto == False:
if args.open_method == "cls EVT":
print(Fore.GREEN + 'adv_param:', args.adv_param)
if args.open_method == "EVT" or args.open_method == "cls EVT":
print(Fore.GREEN + 'EVT_threshold:', args.EVT_threshold)
print(Fore.GREEN + 'EVT_power:', args.EVT_power)
print(Fore.GREEN + 'mix_interval_length:', args.mix_interval_length)
if args.entropy_max_loss:
print()
print(Fore.GREEN + "Using entropy maximization loss ...")
print(Fore.GREEN + 'lambda_:', args.lambda_)
print()
if args.open_method == "OSVM":
print(Fore.GREEN + 'OSVM_threshold:', args.OSVM_threshold)
print(Fore.GREEN + 'Baseline:', args.baseline_type)
print(Fore.GREEN + 'Frame aggregation method:', args.frame_aggregation)
print(Fore.GREEN + 'target data usage:', args.use_target)
if args.use_target == 'none':
print(Fore.GREEN + 'no Domain Adaptation')
else:
if args.dis_DA != 'none':
print(Fore.GREEN + 'Apply the discrepancy-based Domain Adaptation approach:', args.dis_DA)
if len(args.place_dis) != args.add_fc + 2:
raise ValueError(Back.RED + 'len(place_dis) should be equal to add_fc + 2')
if args.adv_DA != 'none':
print(Fore.GREEN + 'Apply the adversarial-based Domain Adaptation approach:', args.adv_DA)
if args.use_bn != 'none':
print(Fore.GREEN + 'Apply the adaptive normalization approach:', args.use_bn)
# determine the categories
class_names = [line.strip().split(' ', 1)[1] for line in open(args.class_file)]
num_class = len(class_names) + 1 if args.open_method == "OSBP" else len(class_names)
class_names.append('UNK')
entropy_cuts = torch.zeros(num_class)
entropy_cuts[entropy_cuts==0] = 100
meter = AverageMeterIvan(num_class) if args.open_method == "OSBP" \
else AverageMeterIvan(num_class + 1)
#=== check the folder existence ===#
path_exp = args.exp_path + args.modality + '/'
if not os.path.isdir(path_exp):
os.makedirs(path_exp)
if args.tensorboard:
writer = SummaryWriter(path_exp + '/tensorboard') # for tensorboardX
#=== initialize the model ===#
if args.auto == False:
print(Fore.CYAN + 'preparing the model......')
model = VideoModel(num_class, args.baseline_type, args.frame_aggregation, args.modality,
train_segments=args.num_segments, val_segments=args.val_segments,
base_model=args.arch, path_pretrained=args.pretrained,
add_fc=args.add_fc, fc_dim = args.fc_dim,
dropout_i=args.dropout_i, dropout_v=args.dropout_v, partial_bn=not args.no_partialbn,
use_bn=args.use_bn if args.use_target != 'none' else 'none', ens_DA=args.ens_DA if args.use_target != 'none' else 'none',
n_rnn=args.n_rnn, rnn_cell=args.rnn_cell, n_directions=args.n_directions, n_ts=args.n_ts,
use_attn=args.use_attn, n_attn=args.n_attn, use_attn_frame=args.use_attn_frame,
verbose=args.verbose, share_params=args.share_params)
model = torch.nn.DataParallel(model).cuda()
if args.optimizer == 'SGD':
# print(Fore.YELLOW + 'using SGD')
optimizer = torch.optim.SGD(model.parameters(), args.lr, momentum=args.momentum, weight_decay=args.weight_decay, nesterov=True)
elif args.optimizer == 'Adam':
# print(Fore.YELLOW + 'using Adam')
optimizer = torch.optim.Adam(model.parameters(), args.lr, weight_decay=args.weight_decay)
else:
print(Back.RED + 'optimizer not support or specified!!!')
exit()
#=== check point ===#
start_epoch = 1
# print(Fore.CYAN + 'checking the checkpoint......')
if args.resume:
if os.path.isfile(args.resume):
checkpoint = torch.load(args.resume)
start_epoch = checkpoint['epoch'] + 1
best_prec1 = checkpoint['best_prec1']
model.load_state_dict(checkpoint['state_dict'])
print(("=> loaded checkpoint '{}' (epoch {})"
.format(args.resume, checkpoint['epoch'])))
if args.resume_hp:
print("=> loaded checkpoint hyper-parameters")
optimizer.load_state_dict(checkpoint['optimizer'])
else:
print(Back.RED + "=> no checkpoint found at '{}'".format(args.resume))
cudnn.benchmark = True
#--- open log files ---#
if not args.evaluate:
if args.resume:
train_file = open(path_exp + 'train.log', 'a')
train_short_file = open(path_exp + 'train_short.log', 'a')
val_file = open(path_exp + 'val.log', 'a')
val_short_file = open(path_exp + 'val_short.log', 'a')
train_file.write('========== start: ' + str(start_epoch) + '\n') # separation line
train_short_file.write('========== start: ' + str(start_epoch) + '\n')
val_file.write('========== start: ' + str(start_epoch) + '\n')
val_short_file.write('========== start: ' + str(start_epoch) + '\n')
else:
train_short_file = open(path_exp + 'train_short.log', 'w')
val_short_file = open(path_exp + 'val_short.log', 'w')
train_file = open(path_exp + 'train.log', 'w')
val_file = open(path_exp + 'val.log', 'w')
val_best_file = open(args.save_best_log, 'a')
else:
test_short_file = open(path_exp + 'test_short.log', 'w')
test_file = open(path_exp + 'test.log', 'w')
#=== Data loading ===#
if args.auto == False:
print(Fore.CYAN + 'loading data......')
if args.use_opencv:
print("use opencv functions")
if args.modality == 'RGB':
data_length = 1
elif args.modality in ['Flow', 'RGBDiff', 'RGBDiff2', 'RGBDiffplus']:
data_length = 5
# calculate the number of videos to load for training in each list ==> make sure the iteration # of source & target are same
num_source = sum(1 for i in open(args.train_source_list))
num_target = sum(1 for i in open(args.train_target_list))
num_val = sum(1 for i in open(args.val_list))
num_iter_source = num_source / args.batch_size[0]
num_iter_target = num_target / args.batch_size[1]
num_max_iter = max(num_iter_source, num_iter_target)
num_source_train = round(num_max_iter*args.batch_size[0]) if args.copy_list[0] == 'Y' else num_source
num_target_train = round(num_max_iter*args.batch_size[1]) if args.copy_list[1] == 'Y' else num_target
# calculate the weight for each class
class_id_list = [int(line.strip().split(' ')[2]) for line in open(args.train_source_list)]
class_id, class_data_counts = np.unique(np.array(class_id_list), return_counts=True)
class_freq = (class_data_counts / class_data_counts.sum()).tolist()
weight_source_class = torch.ones(num_class).cuda()
weight_domain_loss = torch.Tensor([1, 1]).cuda()
if args.weighted_class_loss == 'Y':
weight_source_class = 1 / torch.Tensor(class_freq).cuda()
if args.weighted_class_loss_DA == 'Y':
weight_domain_loss = torch.Tensor([1/num_source_train, 1/num_target_train]).cuda()
# data loading (always need to load the testing data)
val_segments = args.val_segments if args.val_segments > 0 else args.num_segments
val_set = TSNDataSet("", args.val_list, num_dataload=num_val, num_segments=val_segments,
new_length=data_length, modality=args.modality,
image_tmpl="img_{:05d}.t7" if args.modality in ["RGB", "RGBDiff", "RGBDiff2",
"RGBDiffplus"] else args.flow_prefix + "{}_{:05d}.t7",
random_shift=False,
test_mode=True,
)
val_loader = torch.utils.data.DataLoader(val_set, batch_size=args.batch_size[2], shuffle=False,
num_workers=args.workers, pin_memory=True)
if not args.evaluate:
source_set = TSNDataSet("", args.train_source_list, num_dataload=num_source_train, num_segments=args.num_segments,
new_length=data_length, modality=args.modality,
image_tmpl="img_{:05d}.t7" if args.modality in ["RGB", "RGBDiff", "RGBDiff2", "RGBDiffplus"] else args.flow_prefix+"{}_{:05d}.t7",
random_shift=False,
test_mode=True,
)
source_sampler = torch.utils.data.sampler.RandomSampler(source_set)
source_loader = torch.utils.data.DataLoader(source_set, batch_size=args.batch_size[0], shuffle=False, sampler=source_sampler, num_workers=args.workers, pin_memory=True)
target_set = TSNDataSet("", args.train_target_list, num_dataload=num_target_train, num_segments=args.num_segments,
new_length=data_length, modality=args.modality,
image_tmpl="img_{:05d}.t7" if args.modality in ["RGB", "RGBDiff", "RGBDiff2", "RGBDiffplus"] else args.flow_prefix + "{}_{:05d}.t7",
random_shift=False,
test_mode=True,
)
target_sampler = torch.utils.data.sampler.RandomSampler(target_set)
target_loader = torch.utils.data.DataLoader(target_set, batch_size=args.batch_size[1], shuffle=False, sampler=target_sampler, num_workers=args.workers, pin_memory=True)
# --- Optimizer ---#
# define loss function (criterion) and optimizer
if args.loss_type == 'nll':
criterion = torch.nn.CrossEntropyLoss(weight=weight_source_class).cuda()
criterion_domain = torch.nn.CrossEntropyLoss(weight=weight_domain_loss).cuda()
else:
raise ValueError("Unknown loss type")
if args.evaluate:
print(Fore.CYAN + 'evaluation only......')
prec1, entropy_cuts = validate(val_loader, model, criterion, num_class, 0, test_file, meter)
test_short_file.write('%.3f\n' % prec1)
return
#=== Training ===#
start_train = time.time()
if args.auto == False:
print(Fore.CYAN + 'start training......')
beta = args.beta
gamma = args.gamma
mu = args.mu
loss_c_current = 999 # random large number
loss_c_previous = 999 # random large number
attn_source_all = torch.Tensor()
attn_target_all = torch.Tensor()
for epoch in range(start_epoch, args.epochs+1):
## schedule for parameters
alpha = 2 / (1 + math.exp(-1 * (epoch) / args.epochs)) - 1 if args.alpha < 0 else args.alpha
## schedule for learning rate
if args.lr_adaptive == 'loss':
adjust_learning_rate_loss(optimizer, args.lr_decay, loss_c_current, loss_c_previous, '>')
elif args.lr_adaptive == 'none' and epoch in args.lr_steps:
adjust_learning_rate(optimizer, args.lr_decay)
# train for one epoch
loss_c, attn_epoch_source, attn_epoch_target = train(num_class, source_loader, target_loader, model, criterion, criterion_domain, optimizer, epoch, train_file, train_short_file, alpha, beta, gamma, mu, entropy_cuts, class_names)
if args.save_attention >= 0:
attn_source_all = torch.cat((attn_source_all, attn_epoch_source.unsqueeze(0))) # save the attention values
attn_target_all = torch.cat((attn_target_all, attn_epoch_target.unsqueeze(0))) # save the attention values
# update the recorded loss_c
loss_c_previous = loss_c_current
loss_c_current = loss_c
# evaluate on validation set
if epoch % args.eval_freq == 0 or epoch == args.epochs:
prec1, entropy_cuts = validate(val_loader, model, criterion, num_class, epoch, val_file, meter)
# auto
trial.report(prec1, epoch)
# Handle pruning based on the intermediate value.
if trial.should_prune():
raise optuna.exceptions.TrialPruned()
# remember best prec@1 and save checkpoint
is_best = prec1 > best_prec1
if args.auto == False:
line_update = ' ==> updating the best accuracy' if is_best else ''
line_best = "Best score {} vs current score {}".format(best_prec1, prec1) + line_update
print(Fore.YELLOW + line_best)
val_short_file.write('%.3f\n' % prec1)
best_prec1 = max(prec1, best_prec1)
if epoch >= 20 and best_prec1 < 0.7:
break
if epoch >= 30 and best_prec1 - prec1 > 0.2:
break
if args.tensorboard:
writer.add_text('Best_Accuracy', str(best_prec1), epoch)
if args.save_model:
save_checkpoint({
'epoch': epoch,
'arch': args.arch,
'state_dict': model.state_dict(),
'optimizer' : optimizer.state_dict(),
'best_prec1': best_prec1,
'prec1': prec1,
}, is_best, path_exp)
end_train = time.time()
# print(Fore.CYAN + 'total training time:', end_train - start_train)
val_best_file.write('%.3f\n' % best_prec1)
# --- write the total time to log files ---#
line_time = 'total time: {:.3f} '.format(end_train - start_train)
if not args.evaluate:
train_file.write(line_time)
train_short_file.write(line_time)
val_file.write(line_time)
val_short_file.write(line_time)
else:
test_file.write(line_time)
test_short_file.write(line_time)
#--- close log files ---#
if not args.evaluate:
train_file.close()
train_short_file.close()
val_file.close()
val_short_file.close()
else:
test_file.close()
test_short_file.close()
if args.tensorboard:
writer.close()
if args.save_attention >= 0:
np.savetxt('attn_source_' + str(args.save_attention) + '.log', attn_source_all.cpu().detach().numpy(), fmt="%s")
np.savetxt('attn_target_' + str(args.save_attention) + '.log', attn_target_all.cpu().detach().numpy(), fmt="%s")
# auto
return best_prec1
def train(num_class, source_loader, target_loader, model, criterion, criterion_domain, optimizer, epoch, log, log_short, alpha, beta, gamma, mu, entropy_cuts, class_names):
batch_time = AverageMeter()
data_time = AverageMeter()
losses_a = AverageMeter() # adversarial loss
losses_d = AverageMeter() # discrepancy loss
losses_e = AverageMeter() # entropy loss
losses_s = AverageMeter() # ensemble loss
losses_c = AverageMeter() # classification loss
losses_h = AverageMeter() # Entropy Maximization loss
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
if args.no_partialbn:
model.module.partialBN(False)
else:
model.module.partialBN(True)
# switch to train mode
model.train()
end = time.time()
data_loader = enumerate(zip(source_loader, target_loader))
# step info
start_steps = epoch * len(source_loader)
total_steps = args.epochs * len(source_loader)
# initialize the embedding
if args.tensorboard:
feat_source_display = None
label_source_display = None
label_source_domain_display = None
feat_target_display = None
label_target_display = None
label_target_domain_display = None
attn_epoch_source = torch.Tensor()
attn_epoch_target = torch.Tensor()
if args.tsne:
feat_source_list = []
feat_target_list = []
source_label_list = []
target_label_list = []
for i, ((source_data, source_record),(target_data, target_record)) in data_loader:
source_label, target_label = torch.tensor([int(i) for i in source_record[2]]), torch.FloatTensor([int(i) for i in target_record[2]])
# setup hyperparameters
p = float(i + start_steps) / total_steps
beta_dann = 2. / (1. + np.exp(-10 * p)) - 1
beta = [beta_dann if beta[i] < 0 else beta[i] for i in range(len(beta))] # replace the default beta if value < 0
source_size_ori = source_data.size() # original shape
target_size_ori = target_data.size() # original shape
batch_source_ori = source_size_ori[0]
batch_target_ori = target_size_ori[0]
# add dummy tensors to keep the same batch size for each epoch (for the last epoch)
if batch_source_ori < args.batch_size[0]:
source_data_dummy = torch.zeros(args.batch_size[0] - batch_source_ori, source_size_ori[1], source_size_ori[2])
source_data = torch.cat((source_data, source_data_dummy))
if batch_target_ori < args.batch_size[1]:
target_data_dummy = torch.zeros(args.batch_size[1] - batch_target_ori, target_size_ori[1], target_size_ori[2])
target_data = torch.cat((target_data, target_data_dummy))
# add dummy tensors to make sure batch size can be divided by gpu #
if source_data.size(0) % gpu_count != 0:
source_data_dummy = torch.zeros(gpu_count - source_data.size(0) % gpu_count, source_data.size(1), source_data.size(2))
source_data = torch.cat((source_data, source_data_dummy))
if target_data.size(0) % gpu_count != 0:
target_data_dummy = torch.zeros(gpu_count - target_data.size(0) % gpu_count, target_data.size(1), target_data.size(2))
target_data = torch.cat((target_data, target_data_dummy))
# measure data loading time
data_time.update(time.time() - end)
source_label = source_label.cuda(non_blocking=True) # pytorch 0.4.X
target_label = target_label.cuda(non_blocking=True) # pytorch 0.4.X
if args.baseline_type == 'frame':
source_label_frame = source_label.unsqueeze(1).repeat(1,args.num_segments).view(-1) # expand the size for all the frames
target_label_frame = target_label.unsqueeze(1).repeat(1, args.num_segments).view(-1)
label_source = source_label_frame if args.baseline_type == 'frame' else source_label # determine the label for calculating the loss function
label_target = target_label_frame if args.baseline_type == 'frame' else target_label
#====== pre-train source data ======#
if args.pretrain_source:
#------ forward pass data again ------#
_, out_source, out_source_2, _, _, _, _, _, _, _ = model(source_data, target_data, beta, mu, is_train=True, reverse=False)
# ignore dummy tensors
out_source = out_source[:batch_source_ori]
out_source_2 = out_source_2[:batch_source_ori]
#------ calculate the loss function ------#
# 1. calculate the classification loss
out = out_source
label = label_source
loss = criterion(out, label)
if args.ens_DA == 'MCD' and args.use_target != 'none':
loss += criterion(out_source_2, label)
# compute gradient and do SGD step
optimizer.zero_grad()
loss.backward()
if args.clip_gradient is not None:
total_norm = clip_grad_norm_(model.parameters(), args.clip_gradient)
if total_norm > args.clip_gradient and args.verbose:
print("clipping gradient: {} with coef {}".format(total_norm, args.clip_gradient / total_norm))
optimizer.step()
#====== forward pass data ======#
attn_source, out_source, out_source_2, pred_domain_source, feat_source, attn_target, out_target, out_target_2, pred_domain_target, feat_target = model(source_data, target_data, beta, mu, is_train=True, reverse=False)
# ignore dummy tensors
attn_source, out_source, out_source_2, pred_domain_source, feat_source = removeDummy(attn_source, out_source, out_source_2, pred_domain_source, feat_source, batch_source_ori)
attn_target, out_target, out_target_2, pred_domain_target, feat_target = removeDummy(attn_target, out_target, out_target_2, pred_domain_target, feat_target, batch_target_ori)
if args.pred_normalize == 'Y': # use the uncertainly method (in contruction...)
out_source = out_source / out_source.var().log()
out_target = out_target / out_target.var().log()
if args.tsne:
feat_source_list.append(out_source.cpu())
feat_target_list.append(out_target.cpu())
source_label_list.append(source_label.cpu())
target_label_list.append(target_label.cpu())
# store the embedding
if args.tensorboard:
feat_source_display = feat_source[1] if i==0 else torch.cat((feat_source_display, feat_source[1]), 0)
label_source_display = label_source if i==0 else torch.cat((label_source_display, label_source), 0)
label_source_domain_display = torch.zeros(label_source.size(0)) if i==0 else torch.cat((label_source_domain_display, torch.zeros(label_source.size(0))), 0)
feat_target_display = feat_target[1] if i==0 else torch.cat((feat_target_display, feat_target[1]), 0)
label_target_display = label_target if i==0 else torch.cat((label_target_display, label_target), 0)
label_target_domain_display = torch.ones(label_target.size(0)) if i==0 else torch.cat((label_target_domain_display, torch.ones(label_target.size(0))), 0)
#====== calculate the loss function ======#
# 1. calculate the classification loss
out = out_source
label = label_source
if args.use_target == 'Sv':
out = torch.cat((out, out_target))
label = torch.cat((label, label_target))
loss_classification = criterion(out, label)
if args.ens_DA == 'MCD' and args.use_target != 'none':
loss_classification += criterion(out_source_2, label)
losses_c.update(loss_classification.item(), out_source.size(0)) # pytorch 0.4.X
loss = loss_classification
# 2. calculate the loss for DA
# (I) discrepancy-based approach: discrepancy loss
if args.dis_DA != 'none' and args.use_target != 'none':
loss_discrepancy = 0
kernel_muls = [2.0]*2
kernel_nums = [2, 5]
fix_sigma_list = [None]*2
if args.dis_DA == 'JAN':
# ignore the features from shared layers
feat_source_sel = feat_source[:-args.add_fc]
feat_target_sel = feat_target[:-args.add_fc]
size_loss = min(feat_source_sel[0].size(0), feat_target_sel[0].size(0)) # choose the smaller number
feat_source_sel = [feat[:size_loss] for feat in feat_source_sel]
feat_target_sel = [feat[:size_loss] for feat in feat_target_sel]
loss_discrepancy += JAN(feat_source_sel, feat_target_sel, kernel_muls=kernel_muls, kernel_nums=kernel_nums, fix_sigma_list=fix_sigma_list, ver=2)
else:
# extend the parameter list for shared layers
kernel_muls.extend([kernel_muls[-1]]*args.add_fc)
kernel_nums.extend([kernel_nums[-1]]*args.add_fc)
fix_sigma_list.extend([fix_sigma_list[-1]]*args.add_fc)
for l in range(0, args.add_fc + 2): # loss from all the features (+2 because of frame-aggregation layer + final fc layer)
if args.place_dis[l] == 'Y':
# select the data for calculating the loss (make sure source # == target #)
size_loss = min(feat_source[l].size(0), feat_target[l].size(0)) # choose the smaller number
# select
feat_source_sel = feat_source[l][:size_loss]
feat_target_sel = feat_target[l][:size_loss]
# break into multiple batches to avoid "out of memory" issue
size_batch = min(256,feat_source_sel.size(0))
feat_source_sel = feat_source_sel.view((-1,size_batch) + feat_source_sel.size()[1:])
feat_target_sel = feat_target_sel.view((-1,size_batch) + feat_target_sel.size()[1:])
if args.dis_DA == 'CORAL':
losses_coral = [CORAL(feat_source_sel[t], feat_target_sel[t]) for t in range(feat_source_sel.size(0))]
loss_coral = sum(losses_coral)/len(losses_coral)
loss_discrepancy += loss_coral
elif args.dis_DA == 'DAN':
losses_mmd = [mmd_rbf(feat_source_sel[t], feat_target_sel[t], kernel_mul=kernel_muls[l], kernel_num=kernel_nums[l], fix_sigma=fix_sigma_list[l], ver=2) for t in range(feat_source_sel.size(0))]
loss_mmd = sum(losses_mmd) / len(losses_mmd)
loss_discrepancy += loss_mmd
else:
raise NameError('not in dis_DA!!!')
losses_d.update(loss_discrepancy.item(), feat_source[0].size(0))
loss += alpha * loss_discrepancy
# (II) adversarial discriminative model: adversarial loss
if args.adv_DA != 'none' and args.use_target != 'none' and args.open_method != 'cls EVT':
loss_adversarial = 0
pred_domain_all = []
pred_domain_target_all = []
for l in range(len(args.place_adv)):
if args.place_adv[l] == 'Y':
# reshape the features (e.g. 128x5x2 --> 640x2)
pred_domain_source_single = pred_domain_source[l].view(-1, pred_domain_source[l].size()[-1])
pred_domain_target_single = pred_domain_target[l].view(-1, pred_domain_target[l].size()[-1])
# prepare domain labels
source_domain_label = torch.zeros(pred_domain_source_single.size(0)).long()
target_domain_label = torch.ones(pred_domain_target_single.size(0)).long()
domain_label = torch.cat((source_domain_label,target_domain_label),0)
domain_label = domain_label.cuda(non_blocking=True)
pred_domain = torch.cat((pred_domain_source_single, pred_domain_target_single),0)
pred_domain_all.append(pred_domain)
pred_domain_target_all.append(pred_domain_target_single)
if args.pred_normalize == 'Y': # use the uncertainly method (in construction......)
pred_domain = pred_domain / pred_domain.var().log()
loss_adversarial_single = criterion_domain(pred_domain, domain_label)
# loss_adversarial_single = ivan_CEL(pred_domain,
# domain_label)
loss_adversarial += loss_adversarial_single
losses_a.update(loss_adversarial.item(), pred_domain.size(0))
loss += loss_adversarial
# (II) adversarial discriminative model: adversarial loss - weighted
if args.adv_DA != 'none' and args.use_target != 'none' and args.open_method == 'cls EVT':
loss_adversarial = 0
pred_domain_all = []
pred_domain_target_all = []
for l in range(len(args.place_adv)):
if args.place_adv[l] == 'Y':
# reshape the features (e.g. 128x5x2 --> 640x2)
pred_domain_source_single = pred_domain_source[l].view(-1, pred_domain_source[l].size()[-1])
pred_domain_target_single = pred_domain_target[l].view(-1, pred_domain_target[l].size()[-1])
# prepare domain labels
source_domain_label = torch.zeros(pred_domain_source_single.size(0)).long()
target_domain_label = torch.ones(pred_domain_target_single.size(0)).long()
domain_label = torch.cat((source_domain_label,target_domain_label),0)
domain_label = domain_label.cuda(non_blocking=True)
pred_domain = torch.cat((pred_domain_source_single, pred_domain_target_single),0)
pred_domain_all.append(pred_domain)
pred_domain_target_all.append(pred_domain_target_single)
if args.pred_normalize == 'Y': # use the uncertainly method (in construction......)
pred_domain = pred_domain / pred_domain.var().log()
loss_adversarial_single = criterion_domain(pred_domain, domain_label)
# loss_adversarial_single = ivan_CEL(pred_domain,
# domain_label)
loss_adversarial_single_source = criterion_domain(pred_domain_source_single, source_domain_label.cuda(non_blocking=True))
#------- Computing Target Advserial Weights -------#
target_entropy = -(F.softmax(out_target, dim=1) * F.log_softmax(out_target, dim=1)).sum(dim=1)
# th-based
# 1. change mininum weight to 0
# 2. remove weight upper bound
weights = []
even_tensor = torch.ones(num_class)/num_class
entropy_max = -(F.softmax(even_tensor, dim=0)* F.log_softmax(even_tensor, dim=0)).sum()
# entropy_cuts[entropy_cuts==100] = -0.5 # 100 means
for i_ in range(list(out_target.shape)[0]):
max_class = int(out_target.data.max(1)[1][i_])
entropy_cut = float(entropy_cuts[max_class])
if entropy_cut > float(entropy_max)/2:
hard_interval_half = (float(entropy_max) - entropy_cut)/2
else:
hard_interval_half = entropy_cut/2
# # use fixed value 0.5 as interval
# weight = entropy_cut - float(target_entropy[i_]) + 0.5
# max = 2.2 t=1.4 half = 0.4
# given h = 1.4: 0.5 + 0.5 * (1.4-1.4)/0.4 = 0.5
# given h = 1.6: 0.5 + 0.5 * (1.4-1.6)/0.4 = 0.25
# given h = 1: 0.5 + 0.5 * (1.4-1)/0.4 = 0
weight = 0.5 + args.mix_interval_length * (entropy_cut - float(target_entropy[i_])) / hard_interval_half # 0.25 0.75 1
weight = 0 if weight < 0 else weight
weight = 1 if weight > 1 else weight
weights.append(weight)
weights = torch.tensor(weights)
# weights = F.normalize(weights.float(), dim=0, p=1) * list(weights.shape)[0]
# entropy-based
# # even_tensor = torch.ones(num_class)/num_class
# # entropy_max = -(F.softmax(even_tensor, dim=0)* F.log_softmax(even_tensor, dim=0)).sum()
# weights = entropy_max - target_entropy
# weights = F.normalize(weights.float(), dim=0, p=1) * list(weights.shape)[0]
if list(weights.shape)[0] != list(pred_domain_target_single.shape)[0]:
weights = tile(weights, 0, args.num_segments)
loss_adversarial_single_target = ivan_CEL(pred_domain_target_single, target_domain_label.cuda(non_blocking=True), weights.cuda())
# loss_adversarial += loss_adversarial_single
loss_adversarial += loss_adversarial_single_source * 0.5
loss_adversarial += loss_adversarial_single_target * 0.5
losses_a.update(loss_adversarial.item(), pred_domain.size(0))
loss += args.adv_param * loss_adversarial
# (III) other loss
# 1. entropy loss for target data
if args.add_loss_DA == 'target_entropy' and args.use_target != 'none':
loss_entropy = cross_entropy_soft(out_target)
losses_e.update(loss_entropy.item(), out_target.size(0))
loss += gamma * loss_entropy
# 2. discrepancy loss for MCD (CVPR 18)
if args.ens_DA == 'MCD' and args.use_target != 'none':
_, _, _, _, _, attn_target, out_target, out_target_2, pred_domain_target, feat_target = model(source_data, target_data, beta, mu, is_train=True, reverse=True)
# ignore dummy tensors
_, out_target, out_target_2, _, _ = removeDummy(attn_target, out_target, out_target_2, pred_domain_target, feat_target, batch_target_ori)
loss_dis = -dis_MCD(out_target, out_target_2)
losses_s.update(loss_dis.item(), out_target.size(0))
loss += loss_dis
# 3. attentive entropy loss
if args.add_loss_DA == 'attentive_entropy' and args.use_attn != 'none' and args.use_target != 'none':
loss_entropy = attentive_entropy(torch.cat((out_source, out_target),0), pred_domain_all[1])
losses_e.update(loss_entropy.item(), out_target.size(0))
loss += gamma * loss_entropy
# 5. H loss
if args.entropy_max_loss == True:
loss_H = H_loss(out_target)
losses_h.update(loss_H.item(), out_target.size(0))
loss += args.lambda_ * loss_H
pred = out
prec1, prec5 = accuracy(pred.data, label, topk=(1, 2))
losses.update(loss.item())
top1.update(prec1.item(), out_source.size(0))
top5.update(prec5.item(), out_source.size(0))
# compute gradient and do SGD step
optimizer.zero_grad()
loss.backward()
if args.clip_gradient is not None:
total_norm = clip_grad_norm_(model.parameters(), args.clip_gradient)
if total_norm > args.clip_gradient and args.verbose:
print("clipping gradient: {} with coef {}".format(total_norm, args.clip_gradient / total_norm))
optimizer.step()
# measure elapsed time
batch_time.update(time.time() - end)
end = time.time()
if i % args.print_freq == 0 and args.auto == False:
line = 'Train: [{0}][{1}/{2}], lr: {lr:.5f}\t' + \
'Time {batch_time.val:.3f} ({batch_time.avg:.3f})\t' + \
'Data {data_time.val:.3f} ({data_time.avg:.3f})\t' + \
'Prec@1 {top1.val:.3f} ({top1.avg:.3f})\t' + \
'Prec@5 {top5.val:.3f} ({top5.avg:.3f})\t' + \
'Loss {loss.val:.4f} ({loss.avg:.4f}) ' \
'loss_c {loss_c.avg:.4f}\t loss_h {loss_h.avg:.4f}\t'
if args.dis_DA != 'none' and args.use_target != 'none':
line += 'alpha {alpha:.3f} loss_d {loss_d.avg:.4f}\t'
if args.adv_DA != 'none' and args.use_target != 'none':
line += 'beta {beta[0]:.3f}, {beta[1]:.3f}, {beta[2]:.3f} loss_a {loss_a.avg:.4f}\t'
if args.add_loss_DA != 'none' and args.use_target != 'none':
line += 'gamma {gamma:.6f} loss_e {loss_e.avg:.4f}\t'
if args.ens_DA != 'none' and args.use_target != 'none':
line += 'mu {mu:.6f} loss_s {loss_s.avg:.4f}\t'
line = line.format(
epoch, i, len(source_loader), batch_time=batch_time, data_time=data_time, alpha=alpha, beta=beta, gamma=gamma, mu=mu,
loss=losses, loss_c=losses_c, loss_h=losses_h, loss_d=losses_d, loss_a=losses_a, loss_e=losses_e, loss_s=losses_s, top1=top1, top5=top5,
lr=optimizer.param_groups[0]['lr'])
if i % args.show_freq == 0:
print(line)
log.write('%s\n' % line)
# adjust the learning rate for ech step (e.g. DANN)
if args.lr_adaptive == 'dann':
adjust_learning_rate_dann(optimizer, p)
# save attention values w/ the selected class
if args.save_attention >= 0:
attn_source = attn_source[source_label==args.save_attention]
attn_target = attn_target[target_label==args.save_attention]
attn_epoch_source = torch.cat((attn_epoch_source, attn_source.cpu()))
attn_epoch_target = torch.cat((attn_epoch_target, attn_target.cpu()))
if args.tsne and epoch >=8 and epoch <= 25:
feat_source = torch.cat(feat_source_list).squeeze()
feat_target = torch.cat(feat_target_list).squeeze()
source_label = torch.cat(source_label_list).squeeze()
target_label = torch.cat(target_label_list).squeeze()
target_label[target_label >= num_class] = num_class
save_tsne_path='tsn/TA3N_'+ str(epoch) +'_.png'
visualize_TSNE(feat_source, feat_target, source_label, target_label, save_tsne_path, class_names)
print('finshed tsne')
# update the embedding every epoch
if args.tensorboard:
n_iter_train = epoch * len(source_loader) # calculate the total iteration
# embedding
# see source and target separately
writer.add_embedding(feat_source_display, metadata=label_source_display.data, global_step=n_iter_train, tag='train_source')
writer.add_embedding(feat_target_display, metadata=label_target_display.data, global_step=n_iter_train, tag='train_target')
# mix source and target
feat_all_display = torch.cat((feat_source_display, feat_target_display), 0)
label_all_domain_display = torch.cat((label_source_domain_display, label_target_domain_display), 0)
writer.add_embedding(feat_all_display, metadata=label_all_domain_display.data, global_step=n_iter_train, tag='train_DA')
# emphazise some classes (1, 3, 11 here)
label_source_1 = 1 * torch.eq(label_source_display, torch.cuda.LongTensor([1]).repeat(label_source_display.size(0))).long().cuda(non_blocking=True)
label_source_3 = 2 * torch.eq(label_source_display, torch.cuda.LongTensor([3]).repeat(label_source_display.size(0))).long().cuda(non_blocking=True)
label_source_11 = 3 * torch.eq(label_source_display, torch.cuda.LongTensor([11]).repeat(label_source_display.size(0))).long().cuda(non_blocking=True)
label_target_1 = 4 * torch.eq(label_target_display, torch.cuda.LongTensor([1]).repeat(label_target_display.size(0))).long().cuda(non_blocking=True)
label_target_3 = 5 * torch.eq(label_target_display, torch.cuda.LongTensor([3]).repeat(label_target_display.size(0))).long().cuda(non_blocking=True)
label_target_11 = 6 * torch.eq(label_target_display, torch.cuda.LongTensor([11]).repeat(label_target_display.size(0))).long().cuda(non_blocking=True)
label_source_display_new = label_source_1 + label_source_3 + label_source_11
id_source_show = ~torch.eq(label_source_display_new, 0).cuda(non_blocking=True)
label_source_display_new = label_source_display_new[id_source_show]
feat_source_display_new = feat_source_display[id_source_show]
label_target_display_new = label_target_1 + label_target_3 + label_target_11
id_target_show = ~torch.eq(label_target_display_new, 0).cuda(non_blocking=True)
label_target_display_new = label_target_display_new[id_target_show]
feat_target_display_new = feat_target_display[id_target_show]
feat_all_display_new = torch.cat((feat_source_display_new, feat_target_display_new), 0)
label_all_display_new = torch.cat((label_source_display_new, label_target_display_new), 0)
writer.add_embedding(feat_all_display_new, metadata=label_all_display_new.data, global_step=n_iter_train, tag='train_DA_labels')
if args.auto == False:
log_short.write('%s\n' % line)
return losses_c.avg, attn_epoch_source.mean(0), attn_epoch_target.mean(0)
def validate(val_loader, model, criterion, num_class, epoch, log, meter):
batch_time = AverageMeter()
losses = AverageMeter()
top1 = AverageMeter()
top5 = AverageMeter()
# switch to evaluate mode
model.eval()
meter.reset()
correct1 = 0
size = 0
labels = []
end = time.time()
target_entropy_all = [] # for global-level EVT
target_logits_cls, target_entropy_cls = [], [] # for class-level EVT
for cls in range(num_class):
target_logits_cls.append([])
target_entropy_cls.append([])
romove_unk = False if epoch != 61 else True
if args.open_method != 'cls EVT':
entropy_cuts = []
# initialize the embedding
if args.tensorboard:
feat_val_display = None
label_val_display = None
for i, (val_data, val_record) in enumerate(val_loader):
val_label = torch.tensor([int(i) for i in val_record[2]])
val_path = val_record[0]
val_size_ori = val_data.size() # original shape
batch_val_ori = val_size_ori[0]
# add dummy tensors to keep the same batch size for each epoch (for the last epoch)