-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy patheval.py
1102 lines (1035 loc) · 56.4 KB
/
eval.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
## python library
import builtins
import math
import os
import json
import random
import shutil
import time
import warnings
import numpy as np
import csv
from sklearn.metrics import confusion_matrix
from PIL import Image, ImageFile
ImageFile.LOAD_TRUNCATED_IMAGES = True
from tqdm import tqdm
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.backends.cudnn as cudnn
import torch.distributed as dist
import torch.optim
import torch.multiprocessing as mp
import torch.utils.data
import torch.utils.data.distributed
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import torch.nn.functional as F
import struct
import io
from DataLoader.example_pb2 import Example
from config_train import parser
from utils.lr_scheduler_webFG import lr_scheduler as lr_scheduler_webFG
from train import init_prototype_fewshot
from model import FoPro, init_weights
from feat_tsne import run_and_plot_tsne
import DataLoader.webFG_dataset as webFG496
import DataLoader.webvision_dataset as webvision
import tensorboard_logger as tb_logger
import warnings
warnings.filterwarnings('ignore')
def get_tfrecord_image(record_file, offset):
"""read images from tfrecord"""
def _parser(feature_list):
"""get the image file and perform transformation
feature_list: the dictionary to load features (images)
"""
for key, feature in feature_list:
if key == 'image':
image_raw = feature.bytes_list.value[0]
image = Image.open(io.BytesIO(image_raw))
image = image.convert('RGB')
return image
return
with open(record_file, 'rb') as ifs:
ifs.seek(offset)
byte_len_crc = ifs.read(12)
proto_len = struct.unpack('Q', byte_len_crc[:8])[0]
pb_data = ifs.read(proto_len)
if len(pb_data) < proto_len:
print("read pb_data err, proto_len:%s pb_data len:%s" % (proto_len, len(pb_data)))
return
example = Example()
example.ParseFromString(pb_data)
# keep key value in order
feature = sorted(example.features.feature.items())
image = _parser(feature)
return image
def main():
args = parser.parse_args()
if args.seed is not None:
np.random.seed(args.seed)
random.seed(args.seed)
torch.manual_seed(args.seed)
torch.cuda.manual_seed(args.seed)
print('You have chosen to seed training.')
# cudnn.deterministic = True
# warnings.warn('This will turn on the CUDNN deterministic setting, '
# 'which can slow down your training considerably! '
# 'You may see unexpected behavior when restarting '
# 'from checkpoints.')
args.gpu = 0
if args.gpu is not None:
warnings.warn('You have chosen a specific GPU. This will completely '
'disable data parallelism.')
args.distributed = False
print("distributed training {}".format(args.distributed))
## prepare the directory for saving
os.makedirs(args.exp_dir, exist_ok=True)
args.tensorboard_dir = os.path.join(args.exp_dir, 'tensorboard')
os.makedirs(args.tensorboard_dir, exist_ok=True)
ngpus_per_node = torch.cuda.device_count()
print("{} single process running with {} gpus".format(args.world_size,\
ngpus_per_node))
## start evaluation inference
main_worker(args.gpu, ngpus_per_node, args)
## start qualitative illustration
if args.webvision:
target_domain_dir = args.root_dir_test_target
target_domain_imglist = args.pathlist_test_target
fast_eval = args.fast_eval
check_statistics_webvision(args.exp_dir, target_domain_dir,\
target_domain_imglist, fast_eval)
else:
target_domain_dir = args.root_dir
target_domain_imglist = os.path.join(target_domain_dir, "val-list.txt")
check_statistics_webFG496(args.exp_dir, target_domain_dir, target_domain_imglist)
run_and_plot_tsne(save_root_path=args.exp_dir)
return
def main_worker(gpu, ngpus_per_node, args):
args.gpu = gpu
if args.gpu is not None:
print("Use GPU: {} for training".format(args.gpu))
# create model
print("=> creating model '{}'".format(args.arch))
model = FoPro(args)
if not (args.pretrained):
model.apply(init_weights)
if args.gpu == 0:
print(model)
torch.cuda.set_device(args.gpu)
model = model.cuda(args.gpu)
## optionally resume from a checkpoint
resume_path = args.resume
assert os.path.exists(resume_path) and os.path.isfile(resume_path)
print("=> loading checkpoint '{}'".format(resume_path))
if args.gpu is None:
checkpoint = torch.load(resume_path)
else:
# Map model to be loaded to specified single gpu.
loc = 'cuda:{}'.format(args.gpu)
checkpoint = torch.load(resume_path, map_location=loc)
state_dict = checkpoint['state_dict']
for k in list(state_dict.keys()):
if k.startswith('module.'):
# remove prefix
state_dict[k.replace('module.','')] = state_dict[k]
del state_dict[k]
model.load_state_dict(state_dict)
epoch = checkpoint['epoch']
print("=> loaded checkpoint '{}' (epoch {})"
.format(resume_path, epoch))
cudnn.benchmark = True
# Data loading code
assert(os.path.exists(args.root_dir)), "please make sure the path to web data is valid {}".format(args.root_dir)
assert(os.path.exists(args.root_dir_t)), "please make sure the path to fewshot target domain data is valid {}".format(args.root_dir_t)
assert(os.path.isfile(args.pathlist_t)), "please make sure the pathlist path to fewshot target domain data is valid {}".format(args.pathlist_t)
if args.webvision:
## load webvision dataset
assert(os.path.isfile(args.pathlist_web)), "please make sure the pathlist path to webvision web data is valid"
assert(os.path.exists(args.root_dir_test_web)), "please make sure the path to webvision web test data is valid"
assert(os.path.isfile(args.pathlist_test_web)), "please make the pathlist path to webvision web test data is valid"
assert(os.path.exists(args.root_dir_test_target)), "please make sure the path to webvision imgnet test data is valid"
assert(os.path.isfile(args.pathlist_test_target)), "please make the pathlist path to webvision imgnet test data is valid"
loader = webvision.webvision_dataloader(batch_size=args.batch_size, num_class=args.num_class, num_workers=args.workers,\
root_dir=args.root_dir, pathlist=args.pathlist_web,\
root_dir_test_web=args.root_dir_test_web,\
pathlist_test_web=args.pathlist_test_web,\
root_dir_test_target=args.root_dir_test_target,\
pathlist_test_target=args.pathlist_test_target, distributed=args.distributed, crop_size=0.2,\
root_dir_target=args.root_dir_t, pathlist_target=args.pathlist_t,\
save_dir=args.exp_dir, dry_run=args.dry_run,\
use_fewshot=args.use_fewshot, fast_eval=args.fast_eval, eval_only=True)
train_loader, fewshot_loader, test_loader_web, test_loader_target = loader.run()
else:
## load webFG496 dataset
loader = webFG496.webFG496_dataloader(batch_size=args.batch_size, num_class=args.num_class,\
num_workers=args.workers, root_dir=args.root_dir, distributed=args.distributed, crop_size=0.2,\
root_dir_target=args.root_dir_t, pathlist_target=args.pathlist_t,\
save_dir=args.exp_dir, dry_run=args.dry_run,\
use_fewshot=args.use_fewshot, eval_only=True)
train_loader, fewshot_loader, test_loader = loader.run()
if args.gpu==0:
logger = tb_logger.Logger(logdir=args.tensorboard_dir, flush_secs=2)
else:
logger = None
if epoch < args.init_proto_epoch:
## 过一遍模型得到fewshot样本作为prototype
if args.use_fewshot:
print("===>Extract features of fewshot samples as prototypes")
init_prototype_fewshot(model, fewshot_loader, args, epoch, is_eval=True)
else:
print("===>Extract features of training samples/average per class as prototypes")
init_prototype_fewshot(model, train_loader, args, epoch, is_eval=True)
else:
print("===>Use model default prototypes")
print("Save model prototypes")
prototypes = model.prototypes
save_path_prototypes = os.path.join(args.exp_dir, "prototypes.npy")
np.save(save_path_prototypes, prototypes.cpu().numpy())
acc1_train, acc5_train = test(model, train_loader, args, 0, logger,\
dataset_name="WebTrain", is_trainset=True)
print("Web Train Accuracy top 1 {} top 5 {}".format(acc1_train, acc5_train))
acc1_fewshot, acc5_fewshot = test(model, fewshot_loader, args, 0, logger, dataset_name="FewShot")
print("Web FewShot Accuracy top 1 {} top 5 {}".format(acc1_fewshot, acc5_fewshot))
if args.webvision:
## test webvision dataset
acc1_web, acc5_web = test(model, test_loader_web, args, 0, logger, dataset_name="WebVision")
print("Acc webvision top 1 {} top 5 {}".format(acc1_web, acc5_web))
acc1_imgnet, acc5_imgnet = test(model, test_loader_target, args, 0, logger, dataset_name="ImgNet")
print("Acc imagenet top 1 {} top 5 {}".format(acc1_imgnet, acc5_imgnet))
else:
## test webFineGrained dataset
acc1, acc5 = test(model, test_loader, args, 0, logger, dataset_name="FineGrained")
print("Acc FGdataset top 1 {} top 5 {}".format(acc1, acc5))
return
def test(model, test_loader, args, epoch, tb_logger, dataset_name="WebVision", is_trainset=False):
with torch.no_grad():
print('==> Evaluation...')
model.eval()
top1_acc = AverageMeter("Top1")
top5_acc = AverageMeter("Top5")
save_path_csv = os.path.join(args.exp_dir, dataset_name+".csv")
save_path_clean_csv = os.path.join(args.exp_dir, dataset_name+"_clean_flag.csv")
csv_files = []
# evaluate on webvision val set
img_features = []
save_path_features = os.path.join(args.exp_dir, dataset_name+"_feat.npy")
img_path_is_clean_record = []
for batch in tqdm(test_loader):
## outputs, feat, target, feat_reconstruct
outputs, q_compress, relation_score, targets_idx, distances = model(batch, args,\
is_analysis=True, is_eval=not (is_trainset), is_proto=True, is_clean=3)
# if is_trainset:
target, dist_min_idx, dist_median_idx, dist_max_idx, clean_idx, arcface_idx = targets_idx
dist_target, dist_pred, dist_min, dist_median, dist_max, dist_sim = distances
# else:
# target, dist_min_idx, dist_median_idx, dist_max_idx, clean_idx = targets_idx
# dist_target, dist_pred, dist_min, dist_median, dist_max = distances
## cosine distance the smaller the closer/more similar to prototype
res, pred_top = accuracy(outputs, target, topk=(1, 5))
relation_score_target = relation_score[target>=0, target]
relation_score_pred = relation_score[pred_top>=0, pred_top]
relation_score_max, rm_max_label = relation_score.max(1)
## img, target, torch.Tensor([self.domains[index]]), path, torch.Tensor([self.samples_index[index]])
if is_trainset:
pathlist = batch[4]
else:
pathlist = batch[3]
# if is_trainset:
for img_path, img_label, img_pred, img_dist_target, img_dist_pred,\
img_dist_min, img_dist_min_id,\
img_dist_median, img_dist_median_id,\
img_dist_max, img_dist_max_id, img_dist_sim in zip(pathlist,\
target, pred_top, dist_target, dist_pred,\
dist_min, dist_min_idx,\
dist_median, dist_median_idx,\
dist_max, dist_max_idx, dist_sim.view(-1)):
csv_files.append([img_path, img_label.item(), img_pred.item(),\
img_dist_target.item(), img_dist_pred.item(),\
img_dist_min.item(), int(img_dist_min_id.item()),\
img_dist_median.item(), int(img_dist_median_id.item()),\
img_dist_max.item(), int(img_dist_max_id.item()),\
float(img_dist_sim.item())])
for img_path, img_label, img_pred,\
img_is_clean, img_is_arcface,\
img_rm_label, img_rm_pred,\
img_rm_max_id, img_rm_max in zip(
pathlist, target, pred_top, clean_idx, arcface_idx,\
relation_score_target.view(-1), relation_score_pred.view(-1),\
rm_max_label.view(-1), relation_score_max.view(-1)):
img_path_is_clean_record.append([img_path,\
int(img_label.item()), int(img_pred.item()),\
int(img_is_clean.item()), int(img_is_arcface.item()),\
float(img_rm_label.item()), float(img_rm_pred.item()),\
int(img_rm_max_id.item()), float(img_rm_max.item())])
# else:
# for img_path, img_label, img_pred, img_dist_target, img_dist_pred,\
# img_dist_min, img_dist_min_id,\
# img_dist_median, img_dist_median_id,\
# img_dist_max, img_dist_max_id in zip(pathlist,\
# target, pred_top, dist_target, dist_pred,\
# dist_min, dist_min_idx,\
# dist_median, dist_median_idx,\
# dist_max, dist_max_idx):
# csv_files.append([img_path, img_label.item(), img_pred.item(),\
# img_dist_target.item(), img_dist_pred.item(),\
# img_dist_min.item(), int(img_dist_min_id.item()),\
# img_dist_median.item(), int(img_dist_median_id.item()),\
# img_dist_max.item(), int(img_dist_max_id.item())])
# for img_path, img_label, img_pred, img_is_clean,\
# img_rm_label, img_rm_pred,\
# img_rm_max_id, img_rm_max in zip (pathlist,\
# target, pred_top, clean_idx,\
# relation_score_target.view(-1), relation_score_pred.view(-1),\
# rm_max_label.view(-1), relation_score_max.view(-1)):
# img_path_is_clean_record.append([img_path,\
# int(img_label.item()), int(img_pred.item()), int(img_is_clean.item()),\
# float(img_rm_label.item()), float(img_rm_pred.item()),\
# int(img_rm_max_id.item()), float(img_rm_max.item())])
if is_trainset:
domain = batch[3]
for img_domain, img_label, img_feat_compress in zip(domain, target,\
q_compress):
img_domain = img_domain.item()
img_label = img_label.item()
img_feat_compress = img_feat_compress.cpu().numpy()
img_features.append([img_domain, img_label] + img_feat_compress.tolist())
acc1, acc5 = res
top1_acc.update(acc1[0])
top5_acc.update(acc5[0])
if is_trainset:
np.save(save_path_features, np.array(img_features))
with open(save_path_csv, "w") as csvfile:
csv_writer = csv.writer(csvfile)
csv_writer.writerow(['Image Path', 'Image Label', 'Image Prediction',\
'Distance To Target Prototype', 'Distance To Predicted Prototype',\
'Min DistanceToPrototype', 'Min DistanceToPrototype ID',\
'Median DistanceToPrototype', 'Median DistanceToPrototype ID',\
'Max DistanceToPrototype', 'Max DistanceToPrototype ID', 'Distance-Similarity'])
for line in csv_files:
csv_writer.writerow(line)
with open(save_path_clean_csv, "w") as csvfile:
csv_writer = csv.writer(csvfile)
# if is_trainset:
csv_writer.writerow(['Image Path', 'Image Label', 'Image Prediction',\
'Is_Clean', 'Is_Arcface-low-dim-Cls',\
'Relation Score (Label)', 'Relation Score (Pred)',\
'Relation Max Class', 'Relation Score (Max)'])
# else:
# csv_writer.writerow(['Image Path', 'Image Label', 'Image Prediction',\
# 'Is_Clean',\
# 'Relation Score (Label)', 'Relation Score (Pred)',\
# 'Relation Max Class', 'Relation Score (Max)'])
for line in img_path_is_clean_record:
csv_writer.writerow(line)
# average across all processes
acc_tensors = torch.Tensor([top1_acc.avg, top5_acc.avg]).cuda(args.gpu)
print('%s Accuracy is %.2f%% (%.2f%%)'%(dataset_name,\
acc_tensors[0], acc_tensors[1]))
if args.gpu ==0:
tb_logger.log_value('{} top1 Acc'.format(dataset_name),\
acc_tensors[0], epoch)
tb_logger.log_value('{} top5 Acc'.format(dataset_name),\
acc_tensors[1], epoch)
return acc_tensors[0].item(), acc_tensors[1].item()
class AverageMeter(object):
"""Computes and stores the average and current value"""
def __init__(self, name, fmt=':f'):
self.name = name
self.fmt = fmt
self.reset()
def reset(self):
self.val = 0
self.avg = 0
self.sum = 0
self.count = 0
def update(self, val, n=1):
self.val = val
self.sum += val * n
self.count += n
self.avg = self.sum / self.count
def __str__(self):
fmtstr = '{name} {val' + self.fmt + '} ({avg' + self.fmt + '})'
return fmtstr.format(**self.__dict__)
class ProgressMeter(object):
def __init__(self, num_batches, meters, prefix=""):
self.batch_fmtstr = self._get_batch_fmtstr(num_batches)
self.meters = meters
self.prefix = prefix
def display(self, batch):
entries = [self.prefix + self.batch_fmtstr.format(batch)]
entries += [str(meter) for meter in self.meters]
print('\t'.join(entries), flush=True)
return
def _get_batch_fmtstr(self, num_batches):
num_digits = len(str(num_batches // 1))
fmt = '{:' + str(num_digits) + 'd}'
return '[' + fmt + '/' + fmt.format(num_batches) + ']'
def accuracy(output, target, topk=(1,)):
"""Computes the accuracy over the k top predictions for the specified values of k"""
with torch.no_grad():
maxk = max(topk)
batch_size = target.size(0)
_, pred = output.topk(maxk, 1, True, True)
_, pred_top = output.topk(1, 1, True, True)
pred = pred.t()
correct = pred.eq(target.view(1, -1).expand_as(pred))
res = []
for k in topk:
correct_k = correct[:k, :].sum(0)
correct_k = correct_k.view(-1).float().sum(0, keepdim=True)
res.append(correct_k.mul_(100.0 / batch_size))
return res, pred_top.view(-1)
def get_concat_h_resize(im1, im2, resample=Image.BILINEAR, resize_big_image=True):
if im1.height == im2.height:
_im1 = im1
_im2 = im2
elif (((im1.height > im2.height) and resize_big_image) or
((im1.height < im2.height) and not resize_big_image)):
_im1 = im1.resize((int(im1.width * im2.height / im1.height), im2.height), resample=resample)
_im2 = im2
else:
_im1 = im1
_im2 = im2.resize((int(im2.width * im1.height / im2.height), im1.height), resample=resample)
dst = Image.new('RGB', (_im1.width + _im2.width, _im1.height))
dst.paste(_im1, (0, 0))
dst.paste(_im2, (_im1.width, 0))
return dst
def get_concat_v_resize(im1, im2, resample=Image.BILINEAR, resize_big_image=True):
if im1.width == im2.width:
_im1 = im1
_im2 = im2
elif (((im1.width > im2.width) and resize_big_image) or
((im1.width < im2.width) and not resize_big_image)):
_im1 = im1.resize((im2.width, int(im1.height * im2.width / im1.width)), resample=resample)
_im2 = im2
else:
_im1 = im1
_im2 = im2.resize((im1.width, int(im2.height * im1.width / im2.width)), resample=resample)
dst = Image.new('RGB', (_im1.width, _im1.height + _im2.height))
dst.paste(_im1, (0, 0))
dst.paste(_im2, (0, _im1.height))
return dst
def check_statistics_webFG496(root_dir, target_domain_dir, target_domain_imglist):
## 读取真实的测试集图像 按类别分类得到参考图
imgs_gt_by_class = {}
with open(target_domain_imglist, "r") as f:
lines = f.readlines()
for line in lines:
info = line.strip().split(": ")
img_path = info[0]
img_label = int(info[1])
if not (img_label in imgs_gt_by_class):
imgs_gt_by_class[img_label] = []
imgs_gt_by_class[img_label].append(os.path.join(target_domain_dir, img_path))
train_csv_path = os.path.join(root_dir, "WebTrain.csv")
fewshot_csv_path = os.path.join(root_dir, "FewShot.csv")
test_csv_path = os.path.join(root_dir, "FineGrained.csv")
def calc_confusion_matrix(csv_path, class_id2class_name=None, is_trainset=False):
y_pred = []
y_true = []
x_path = []
if class_id2class_name is None:
## 重新构建字典来保存映射
class_id2class_name = {}
with open(csv_path, "r") as f:
csv_reader = csv.reader(f)
for idx, line in enumerate(csv_reader):
if idx == 0:
continue
img_path = line[0]
img_label = line[1]
img_label = int(img_label)
class_name = os.path.basename(os.path.dirname(img_path)).replace(" ", "_")
class_id2class_name[img_label] = class_name
print(class_id2class_name.keys())
## path_to_val_folder/001.Black_footed_Albatross/Black_Footed_Albatross_0001_796111.jpg,0,144
imgs_by_class = {}
with open(csv_path, "r") as f:
csv_reader = csv.reader(f)
for idx, line in enumerate(csv_reader):
if idx == 0:
continue
if len(line) == 11:
img_path, img_label, img_pred,\
img_dist_target, img_dist_pred,\
img_dist_min, img_dist_min_id,\
img_dist_median, img_dist_median_id,\
img_dist_max, img_dist_max_id = line
img_dist_sim = -1
else:
img_path, img_label, img_pred,\
img_dist_target, img_dist_pred,\
img_dist_min, img_dist_min_id,\
img_dist_median, img_dist_median_id,\
img_dist_max, img_dist_max_id,\
img_dist_sim = line
img_label = int(img_label)
img_pred = int(img_pred)
img_dist_target = float(img_dist_target)
img_dist_pred = float(img_dist_pred)
class_name_label = class_id2class_name[img_label]
class_name_pred = class_id2class_name[img_pred]
img_dist_min = float(img_dist_min)
class_name_dist_min = class_id2class_name[int(img_dist_min_id)]
img_dist_median = float(img_dist_median)
class_name_dist_median = class_id2class_name[int(img_dist_median_id)]
img_dist_max = float(img_dist_max)
class_name_dist_max = class_id2class_name[int(img_dist_max_id)]
img_dist_sim = float(img_dist_sim)
img_name_info = "dist_GT_%.2f(%s)_Pred_%.2f(%s)_Sim_%.2f_min_%.2f(%s)_max_%.2f_med_%.2f"%(
img_dist_target, class_name_label,\
img_dist_pred, class_name_pred,\
img_dist_sim,\
img_dist_min, class_name_dist_min,\
img_dist_max,\
img_dist_median
)
if not img_label in imgs_by_class:
imgs_by_class[img_label] = []
imgs_by_class[img_label].append([img_path, img_dist_target, img_name_info])
x_path.append(img_path)
y_true.append(img_label)
y_pred.append(img_pred)
imgs_by_class_clean = {}
imgs_by_relation_score = {}
clean_csv_path = csv_path.replace(".csv", "_clean_flag.csv")
assert(os.path.exists(clean_csv_path))
with open(clean_csv_path, "r") as csvfile:
csv_reader = csv.reader(csvfile)
for idx, line in enumerate(csv_reader):
if idx == 0:
continue
if len(line) == 9:
img_path, img_label, img_pred,\
clean_idx, arcface_idx,\
rel_label, rel_pred,\
rel_max_class, rel_max = line
elif len(line) == 8:
img_path, img_label, img_pred,\
clean_idx,\
rel_label, rel_pred,\
rel_max_class, rel_max = line
arcface_idx = 1
else:
raise ValueError("wrong format of csv file lines")
img_label = int(img_label)
img_pred = int(img_pred)
if not img_label in imgs_by_relation_score:
imgs_by_relation_score[img_label] = []
imgs_by_relation_score[img_label].append([img_path, img_label, rel_label,\
img_pred, rel_pred, rel_max_class, rel_max])
if not img_label in imgs_by_class_clean:
imgs_by_class_clean[img_label] = [[],[]]
if clean_idx:
imgs_by_class_clean[img_label][0].append([img_path, img_pred])
if arcface_idx:
imgs_by_class_clean[img_label][1].append([img_path, img_pred])
for img_label in imgs_by_class:
img_pathlist = imgs_by_class[img_label]
img_pathlist_sorted = sorted(img_pathlist, key=lambda x:x[1], reverse=False)
# images are sorted by their distance to the prototype feature
assert(len(img_pathlist_sorted) == len(img_pathlist))
imgs_by_class[img_label] = img_pathlist_sorted
y_pred = np.array(y_pred)
y_true = np.array(y_true)
x_path = np.array(x_path)
## 计算混淆矩阵
cm = confusion_matrix(y_true, y_pred)
cm_rm_self = np.diag(np.ones(shape=cm.shape[0]) * (np.inf) * (-1)) + cm
## by definition cm_ij is equal to the number of observations
## known to be in group i and predicted to be in group j
## 首先排除对角线上的每个元素类
cm_total_err = np.sum(cm, axis=1) - np.diag(cm)
## 计算每个类别预测成最多的错误类
cm_top_err = np.argmax(cm_rm_self, axis=1)
## 统计该种错误出现的次数confusion_matrix_ij
cm_top_err_num = np.max(cm_rm_self, axis=1)
save_csv_path = os.path.join(os.path.dirname(csv_path), "statistics_CM_" + os.path.basename(csv_path))
with open(save_csv_path, "w") as f:
csv_writer = csv.writer(f)
csv_writer.writerow(["Confusion Matrix"])
top_row = ['GT\Pred']
for i in range(cm.shape[0]):
top_row.append(class_id2class_name[i])
csv_writer.writerow(top_row)
for i in range(cm.shape[0]):
row_i = [class_id2class_name[i]]
for j in range(len(cm[i])):
row_i.append(cm[i][j])
csv_writer.writerow(row_i)
save_csv_path = os.path.join(os.path.dirname(csv_path), "statistics_top_" + os.path.basename(csv_path))
with open(save_csv_path, "w") as f:
csv_writer = csv.writer(f)
csv_writer.writerow(["GT id", "GT Name", "Total Error", "Pred Top 1 Error id", "Pred Top 1 Error id", "Pred Top 1 Error Num"])
for i in range(len(cm_top_err)):
pred_err_id = cm_top_err[i]
row_i = [i, class_id2class_name[i], cm_total_err[i],\
pred_err_id, class_id2class_name[pred_err_id],\
cm_top_err_num[pred_err_id]]
csv_writer.writerow(row_i)
## 统计出现次数最多的错误类别
## 从错误数目最高的前10类图像中抽取每类最容易预测错误的类别图像
## 以及每个数据集前20个类别的样本
top_10_error_class = np.argsort(cm_top_err_num)[::-1][:10]
if is_trainset:
top_20_label_class = np.arange(20)
top_classes = top_10_error_class.tolist() + top_20_label_class.tolist()
top_classes = list(set(top_classes))
else:
top_classes = top_10_error_class
## 且给出每个类别对应图片靠prototype最近的前20张图以及最远的后20张图；
## 以及对应每个类别随机抽20张样本看整体分布情况
## 每张图都匹配一张真实的测试集样本作为参考
save_img_path = os.path.join(os.path.dirname(csv_path), os.path.basename(csv_path).replace(".csv", ""))
os.makedirs(save_img_path, exist_ok=True)
for class_i in top_classes:
## 当前类别名
class_name_i = class_id2class_name[class_i]
## 类别i很容易被模型预测错误的类别
pred_err_i = cm_top_err[class_i]
## 类别i被预测错成其他类别的数目
pred_err_i_num = cm_top_err_num[class_i]
## 预测错的类别名
pred_err_name_i = class_id2class_name[pred_err_i]
print("current class {} is wrongly predicted as class {} with {} times".format(
class_name_i, pred_err_name_i, pred_err_i_num))
## 筛选出所有类别为class_i且被预测成pred_err_i的图像列表
is_class_i = (y_true == class_i)
is_pred_i = (y_pred == pred_err_i)
is_selected = (is_class_i & is_pred_i)
## 首先从当前类别的图像中抽取典型图像
imgs_class_i = imgs_by_class[class_i]
save_img_class_i_path = os.path.join(save_img_path, class_name_i)
os.makedirs(save_img_class_i_path, exist_ok=True)
## 然后从测试集类别图像中抽取正确图像
imgs_gt_class_i = imgs_gt_by_class[class_i]
## 同样从测试集类别图像中抽取错误预测类别的图像
imgs_err_class_i = imgs_gt_by_class[pred_err_i]
img_dist_by_path = {x[0]:x[2] for x in imgs_class_i}
## 存储距离近的一些结果(10)
N_top = 20
for idx, img_item in enumerate(imgs_class_i[:N_top]):
img_path, img_dist, save_name = img_item
im = Image.open(img_path).convert("RGB")
im_gt = Image.open(random.choice(imgs_gt_class_i)).convert("RGB")
im_concat = get_concat_h_resize(im, im_gt)
save_img_class_i_dist_path = os.path.join(save_img_class_i_path,
"%s_idx_%d_R(test_gt).jpg"%(save_name, idx))
im_concat.save(save_img_class_i_dist_path)
## 存储距离远的一些结果(-10)
for idx, img_item in enumerate(imgs_class_i[-N_top:]):
img_path, img_dist, save_name = img_item
im = Image.open(img_path).convert("RGB")
im_gt = Image.open(random.choice(imgs_gt_class_i)).convert("RGB")
im_concat = get_concat_h_resize(im, im_gt)
save_img_class_i_dist_path = os.path.join(save_img_class_i_path,
"%s_idx_%d_R(test_gt).jpg"%(save_name, N_top+idx))
im_concat.save(save_img_class_i_dist_path)
## 筛选对应预测错的的类别以及其对应prototype距离
save_img_class_i_wrong_path = os.path.join(save_img_class_i_path, "errors")
os.makedirs(save_img_class_i_wrong_path, exist_ok=True)
if np.sum(is_selected) > 0:
x_path_selected = x_path[is_selected]
for idx, img_path in enumerate(x_path_selected[:N_top]):
dist_save_name = img_dist_by_path[img_path]
save_img_name_i = "L(GT)_R(Pred)_{}_{}.jpg".format(dist_save_name, idx)
im = Image.open(img_path).convert("RGB")
im_pred_gt = Image.open(random.choice(imgs_err_class_i)).convert("RGB")
im_concat = get_concat_h_resize(im, im_pred_gt)
im_concat.save(os.path.join(save_img_class_i_wrong_path, save_img_name_i))
else:
print("not enough images of class {} wrongly predicted as {}".format(class_name_i,\
pred_err_name_i))
## 存储relation module的输出结果
save_img_class_i_relation_path = os.path.join(save_img_class_i_path, "relation_samples")
os.makedirs(save_img_class_i_relation_path, exist_ok=True)
imgs_all_relation_scores = imgs_by_relation_score[class_i]
imgs_all_relation_scores = sorted(imgs_all_relation_scores, key=lambda x:x[2])
relation_samples = imgs_all_relation_scores[:20] + imgs_all_relation_scores[-20:]
for idx, sample_item in enumerate(relation_samples):
img_path, img_label, rel_label, img_pred, rel_pred, rel_max_class, rel_max = sample_item
save_img_name_i = "{}_GT_{}({})_Pred_{}({})_Max_{}({}).jpg".format(idx,\
rel_label, class_name_i,\
rel_pred, class_id2class_name[img_pred],\
rel_max, class_id2class_name[int(rel_max_class)])
shutil.copyfile(img_path, os.path.join(save_img_class_i_relation_path, save_img_name_i))
## 存储被模型认为是干净样本和参与arcface classifier更新的样本
save_img_class_i_clean_path = os.path.join(save_img_class_i_path, "clean_samples")
save_img_class_i_arcface_path = os.path.join(save_img_class_i_path, "arcface_low-dim_cls_samples")
os.makedirs(save_img_class_i_clean_path, exist_ok=True)
os.makedirs(save_img_class_i_arcface_path, exist_ok=True)
samples_clean_class_i = imgs_by_class_clean[class_i][0]
samples_arcface_class_i = imgs_by_class_clean[class_i][1]
samples_clean_class_i = random.sample(samples_clean_class_i, min(20, len(samples_clean_class_i)))
samples_arcface_class_i = random.sample(samples_arcface_class_i, min(20, len(samples_arcface_class_i)))
for idx, sample_clean in enumerate(samples_clean_class_i):
img_path_idx, img_pred_idx = sample_clean
save_img_name_i = "L(GT)_R(Pred)_{}.jpg".format(idx)
im = Image.open(img_path_idx).convert("RGB")
im_pred_gt = Image.open(random.choice(imgs_gt_by_class[img_pred_idx])).convert("RGB")
im_concat = get_concat_h_resize(im, im_pred_gt)
im_concat.save(os.path.join(save_img_class_i_clean_path, save_img_name_i))
for idx, sample_clean in enumerate(samples_arcface_class_i):
img_path_idx, img_pred_idx = sample_clean
save_img_name_i = "L(GT)_R(Pred)_{}.jpg".format(idx)
im = Image.open(img_path_idx).convert("RGB")
im_pred_gt = Image.open(random.choice(imgs_gt_by_class[img_pred_idx])).convert("RGB")
im_concat = get_concat_h_resize(im, im_pred_gt)
im_concat.save(os.path.join(save_img_class_i_arcface_path, save_img_name_i))
return class_id2class_name
print("================Testing Samples================")
class_id2class_name = calc_confusion_matrix(test_csv_path, class_id2class_name=None, is_trainset=False)
## write mapping from class id to class name
save_path_mapping = os.path.join(root_dir, "class_id2name.txt")
with open(save_path_mapping, "w") as f_write:
for class_id, class_name in class_id2class_name.items():
class_id_name_full = "{}@{}\n".format(class_id, class_name)
f_write.write(class_id_name_full)
print("================FewShot Samples================")
_ = calc_confusion_matrix(fewshot_csv_path, class_id2class_name=class_id2class_name, is_trainset=True)
print("================Training Samples================")
_ = calc_confusion_matrix(train_csv_path, class_id2class_name=class_id2class_name, is_trainset=True)
return
def read_json(json_path_list):
"""read json file and return data"""
with open(json_path_list) as json_file:
json_dict = json.load(json_file)
return json_dict
def check_statistics_webvision(root_dir, target_domain_dir,\
target_domain_imglist, fast_eval=0):
## 读取真实的测试集图像 按类别分类得到参考图
imgs_gt_by_class = {}
with open(target_domain_imglist, "r") as f:
lines = f.readlines()
for line in lines:
info = line.strip().split(" ")
img_path = info[0]
img_label = int(info[1])
if not (img_label in imgs_gt_by_class):
imgs_gt_by_class[img_label] = []
imgs_gt_by_class[img_label].append(os.path.join(target_domain_dir, img_path))
train_csv_path = os.path.join(root_dir, "WebTrain.csv")
fewshot_csv_path = os.path.join(root_dir, "FewShot.csv")
test_web_csv_path = os.path.join(root_dir, "WebVision.csv")
test_img_csv_path = os.path.join(root_dir, "ImgNet.csv")
def calc_confusion_matrix(csv_path, class_id2class_name=None, is_trainset=False):
y_pred = []
y_true = []
x_path = []
if class_id2class_name is None:
## 重新构建字典来保存映射
class_id2class_name = {}
if fast_eval == 2:
class_id2wdnet_id_path = "../imglists/mapping_google_500.txt"
else:
class_id2wdnet_id_path = "../imglists/mapping_webvision_1k.txt"
imgnet_info = "../imagenet_class_1k_full_index.json"
_, _, img_by_wdnet = read_json(imgnet_info)
with open(class_id2wdnet_id_path, "r") as f:
lines = f.readlines()
for line in lines:
info = line.strip().split(" ")
class_id = int(info[0])
wdnet_id = info[1]
img_info = img_by_wdnet[wdnet_id]
if len(img_info) > 2:
class_name = img_info[2].replace(" ", "_")
else:
class_name = img_info[0].replace(" ", "_")
class_id2class_name[class_id] = class_name
## path_to_val_folder/001.Black_footed_Albatross/Black_Footed_Albatross_0001_796111.jpg,0,144
imgs_by_class = {}
with open(csv_path, "r") as f:
csv_reader = csv.reader(f)
for idx, line in enumerate(csv_reader):
if idx == 0:
continue
if len(line) == 11:
img_path, img_label, img_pred,\
img_dist_target, img_dist_pred,\
img_dist_min, img_dist_min_id,\
img_dist_median, img_dist_median_id,\
img_dist_max, img_dist_max_id = line
img_dist_sim = -1
else:
img_path, img_label, img_pred,\
img_dist_target, img_dist_pred,\
img_dist_min, img_dist_min_id,\
img_dist_median, img_dist_median_id,\
img_dist_max, img_dist_max_id,\
img_dist_sim = line
img_label = int(img_label)
img_pred = int(img_pred)
img_dist_target = float(img_dist_target)
img_dist_pred = float(img_dist_pred)
class_name_label = class_id2class_name[img_label]
class_name_pred = class_id2class_name[img_pred]
img_dist_min = float(img_dist_min)
class_name_dist_min = class_id2class_name[int(img_dist_min_id)]
img_dist_median = float(img_dist_median)
class_name_dist_median = class_id2class_name[int(img_dist_median_id)]
img_dist_max = float(img_dist_max)
class_name_dist_max = class_id2class_name[int(img_dist_max_id)]
img_name_info = "dist_GT_%.2f(%s)_Pred_%.2f(%s)_Sim_%.2f_min_%.2f(%s)_max_%.2f_med_%.2f"%(
img_dist_target, class_name_label,\
img_dist_pred, class_name_pred,\
float(img_dist_sim),\
img_dist_min, class_name_dist_min,\
img_dist_max,\
img_dist_median
)
if not img_label in imgs_by_class:
imgs_by_class[img_label] = []
imgs_by_class[img_label].append([img_path, img_dist_target, img_name_info])
x_path.append(img_path)
y_true.append(img_label)
y_pred.append(img_pred)
imgs_by_class_clean = {}
imgs_by_relation_score = {}
clean_csv_path = csv_path.replace(".csv", "_clean_flag.csv")
assert(os.path.exists(clean_csv_path))
with open(clean_csv_path, "r") as csvfile:
csv_reader = csv.reader(csvfile)
for idx, line in enumerate(csv_reader):
if idx == 0:
continue
if len(line) == 9:
img_path, img_label, img_pred,\
clean_idx, arcface_idx,\
rel_label, rel_pred,\
rel_max_class, rel_max = line
elif len(line) == 8:
img_path, img_label, img_pred,\
clean_idx,\
rel_label, rel_pred,\
rel_max_class, rel_max = line
arcface_idx = 1
else:
raise ValueError("wrong format of csv file lines")
img_label = int(img_label)
img_pred = int(img_pred)
if not img_label in imgs_by_relation_score:
imgs_by_relation_score[img_label] = []
imgs_by_relation_score[img_label].append([img_path, img_label, rel_label,\
img_pred, rel_pred, rel_max_class, rel_max])
if not img_label in imgs_by_class_clean:
imgs_by_class_clean[img_label] = [[],[]]
if clean_idx:
imgs_by_class_clean[img_label][0].append([img_path, img_pred])
if arcface_idx:
imgs_by_class_clean[img_label][1].append([img_path, img_pred])
for img_label in imgs_by_class:
img_pathlist = imgs_by_class[img_label]
img_pathlist_sorted = sorted(img_pathlist, key=lambda x:x[1], reverse=False)
# images are sorted by their distance to the prototype feature
assert(len(img_pathlist_sorted) == len(img_pathlist))
imgs_by_class[img_label] = img_pathlist_sorted
y_pred = np.array(y_pred)
y_true = np.array(y_true)
x_path = np.array(x_path)
## 计算混淆矩阵
if fast_eval == 2:
cm = confusion_matrix(y_true, y_pred, labels=np.arange(500))
else:
cm = confusion_matrix(y_true, y_pred, labels=np.arange(1000))
cm_rm_self = np.diag(np.ones(shape=cm.shape[0]) * (np.inf) * (-1)) + cm
## by definition cm_ij is equal to the number of observations
## known to be in group i and predicted to be in group j
## 首先排除对角线上的每个元素类
cm_total_err = np.sum(cm, axis=1) - np.diag(cm)
## 计算每个类别预测成最多的错误类
cm_top_err = np.argmax(cm_rm_self, axis=1)
## 统计该种错误出现的次数confusion_matrix_ij
cm_top_err_num = np.max(cm_rm_self, axis=1)
save_csv_path = os.path.join(os.path.dirname(csv_path), "statistics_CM_" + os.path.basename(csv_path))
with open(save_csv_path, "w") as f:
csv_writer = csv.writer(f)
csv_writer.writerow(["Confusion Matrix"])
top_row = ['GT\Pred']
for i in range(cm.shape[0]):
top_row.append(class_id2class_name[i])
csv_writer.writerow(top_row)
for i in range(cm.shape[0]):
row_i = [class_id2class_name[i]]
for j in range(len(cm[i])):
row_i.append(cm[i][j])
csv_writer.writerow(row_i)
save_csv_path = os.path.join(os.path.dirname(csv_path), "statistics_top_" + os.path.basename(csv_path))
with open(save_csv_path, "w") as f:
csv_writer = csv.writer(f)
csv_writer.writerow(["GT id", "GT Name", "Total Error", "Pred Top 1 Error id", "Pred Top 1 Error id", "Pred Top 1 Error Num"])
for i in range(len(cm_top_err)):
pred_err_id = cm_top_err[i]
row_i = [i, class_id2class_name[i], cm_total_err[i],\
pred_err_id, class_id2class_name[pred_err_id],\
cm_top_err_num[pred_err_id]]
csv_writer.writerow(row_i)
## 统计出现次数最多的错误类别
## 从错误数目最高的前10类图像中抽取每类最容易预测错误的类别图像
## 以及每个数据集前20个类别的样本 + 歧义类别(鼓槌\手推车\鸟\起重机\钉子)
top_10_error_class = np.argsort(cm_top_err_num)[::-1][:10]
class_must_include = []
if fast_eval != 2:
### webvision
class_must_include = [542, 428, 134, 517, 677, 998]
top_10_error_class = top_10_error_class.tolist() + class_must_include
else:
### google500
class_must_include = [137, 246, 385, 227]
top_10_error_class = top_10_error_class.tolist() + class_must_include
if is_trainset:
top_20_label_class = np.arange(20)
top_classes = top_10_error_class + top_20_label_class.tolist()
top_classes = list(set(top_classes))
else:
top_classes = top_10_error_class
## 且给出每个类别对应图片靠prototype最近的前20张图以及最远的后20张图；
## 以及对应每个类别随机抽20张样本看整体分布情况
## 每张图都匹配一张真实的测试集样本作为参考
save_img_path = os.path.join(os.path.dirname(csv_path), os.path.basename(csv_path).replace(".csv", ""))
os.makedirs(save_img_path, exist_ok=True)
for class_i in top_classes:
## 当前类别名
class_name_i = str(class_i) + "_" + class_id2class_name[class_i]
## 类别i很容易被模型预测错误的类别
pred_err_i = cm_top_err[class_i]
## 类别i被预测错成其他类别的数目
pred_err_i_num = cm_top_err_num[class_i]
## 预测错的类别名
pred_err_name_i = class_id2class_name[pred_err_i]
print("current class {} is wrongly predicted as class {} with {} times".format(
class_name_i, pred_err_name_i, pred_err_i_num))
## 筛选出所有类别为class_i且被预测成pred_err_i的图像列表
is_class_i = (y_true == class_i)
is_pred_i = (y_pred == pred_err_i)
is_selected = (is_class_i & is_pred_i)
## 首先从当前类别的图像中抽取典型图像
imgs_class_i = imgs_by_class[class_i]
save_img_class_i_path = os.path.join(save_img_path, class_name_i)
os.makedirs(save_img_class_i_path, exist_ok=True)
## 然后从测试集类别图像中抽取正确图像
imgs_gt_class_i = imgs_gt_by_class[class_i]
## 同样从测试集类别图像中抽取错误预测类别的图像
imgs_err_class_i = imgs_gt_by_class[pred_err_i]
img_dist_by_path = {x[0]:x[2] for x in imgs_class_i}
## 存储距离近的一些结果(10)
if class_i in class_must_include:
print("visiting class {} with all instances".format(class_i))
N_top = len(imgs_class_i)//2
else:
N_top = 40
for idx, img_item in enumerate(imgs_class_i[:N_top]):
img_path, img_dist, save_name = img_item
tf_path, tf_offset = img_path.split("@")
im = get_tfrecord_image(tf_path, int(tf_offset)).convert("RGB")
gt_path = random.choice(imgs_gt_class_i)
tf_path, tf_offset = gt_path.split("@")
im_gt = get_tfrecord_image(tf_path, int(tf_offset)).convert("RGB")
im_concat = get_concat_h_resize(im, im_gt)
save_img_class_i_dist_path = os.path.join(save_img_class_i_path,
"%s_idx_%d_R(test_gt).jpg"%(save_name, idx))
im_concat.save(save_img_class_i_dist_path)
## 存储距离远的一些结果(-10)
for idx, img_item in enumerate(imgs_class_i[-N_top:]):
img_path, img_dist, save_name = img_item