-
Notifications
You must be signed in to change notification settings - Fork 258
/
Copy pathtorchvision_tutorial.html
1267 lines (1042 loc) ยท 110 KB
/
torchvision_tutorial.html
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
<!DOCTYPE html>
<!--[if IE 8]><html class="no-js lt-ie9" lang="ko" > <![endif]-->
<!--[if gt IE 8]><!--> <html class="no-js" lang="ko" > <!--<![endif]-->
<head>
<meta charset="utf-8">
<meta property="og:title" content="TorchVision Object Detection Finetuning Tutorial" />
<meta property="og:type" content="article" />
<meta property="og:url" content="https://tutorials.pytorch.kr/intermediate/torchvision_tutorial.html" />
<meta property="og:site_name" content="PyTorch Tutorials KR" />
<meta property="og:description" content="For this tutorial, we will be finetuning a pre-trained Mask R-CNN model on the Penn-Fudan Database for Pedestrian Detection and Segmentation. It contains 170 images with 345 instances of pedestrians, and we will use it to illustrate how to use the new features in torchvision in order to train an ..." />
<meta property="og:image" content="https://tutorials.pytorch.kr/_static/logos/logo-kr-sm-dark.png" />
<meta property="og:image:alt" content="PyTorch Tutorials KR" />
<meta name="description" content="For this tutorial, we will be finetuning a pre-trained Mask R-CNN model on the Penn-Fudan Database for Pedestrian Detection and Segmentation. It contains 170 images with 345 instances of pedestrians, and we will use it to illustrate how to use the new features in torchvision in order to train an ..." />
<meta property="og:ignore_canonical" content="true" />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>TorchVision Object Detection Finetuning Tutorial — ํ์ดํ ์น ํ๊ตญ์ด ํํ ๋ฆฌ์ผ (PyTorch tutorials in Korean)</title>
<link rel="shortcut icon" href="../_static/favicon.ico"/>
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<!-- <link rel="stylesheet" href="../_static/pygments.css" type="text/css" /> -->
<link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="../_static/copybutton.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.16.10/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="../_static/katex-math.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-binder.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-dataframe.css" type="text/css" />
<link rel="stylesheet" href="../_static/sg_gallery-rendered-html.css" type="text/css" />
<link rel="stylesheet" href="../_static/sphinx-design.5ea377869091fd0449014c60fc090103.min.css" type="text/css" />
<link rel="stylesheet" href="https://cdn.jsdelivr.net/npm/katex@0.10.0-beta/dist/katex.min.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/custom.css" type="text/css" />
<link rel="stylesheet" href="../_static/css/custom2.css" type="text/css" />
<link rel="index" title="์์ธ" href="../genindex.html" />
<link rel="search" title="๊ฒ์" href="../search.html" />
<link rel="next" title="์ปดํจํฐ ๋น์ (Vision)์ ์ํ ์ ์ดํ์ต(Transfer Learning)" href="../beginner/transfer_learning_tutorial.html" />
<link rel="prev" title="TensorBoard๋ก ๋ชจ๋ธ, ๋ฐ์ดํฐ, ํ์ต ์๊ฐํํ๊ธฐ" href="tensorboard_tutorial.html" />
<script src="../_static/js/modernizr.min.js"></script>
<!-- Preload the theme fonts -->
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-book.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-Medium.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/FreightSans/freight-sans-medium-italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="../_static/fonts/IBMPlexMono/IBMPlexMono-SemiBold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<!-- Preload the katex fonts -->
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Math-Italic.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Main-Bold.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size1-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size4-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size2-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Size3-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="preload" href="https://cdn.jsdelivr.net/npm/katex@0.10.0/dist/fonts/KaTeX_Caligraphic-Regular.woff2" as="font" type="font/woff2" crossorigin="anonymous">
<link rel="stylesheet" href="https://use.fontawesome.com/releases/v5.15.2/css/all.css" integrity="sha384-vSIIfh2YWi9wW0r9iZe7RJPrKwp6bG+s9QZMoITbCckVJqGCCRhc+ccxNcdpHuYu" crossorigin="anonymous">
</head>
<div class="container-fluid header-holder tutorials-header" id="header-holder">
<div class="container">
<div class="header-container">
<a class="header-logo" href="https://pytorch.kr/" aria-label="PyTorch"></a>
<div class="main-menu">
<ul>
<li>
<a href="https://pytorch.kr/get-started">์์ํ๊ธฐ</a>
</li>
<li class="active">
<a href="https://tutorials.pytorch.kr/">ํํ ๋ฆฌ์ผ</a>
</li>
<li>
<a href="https://pytorch.kr/hub">ํ๋ธ</a>
</li>
<li>
<a href="https://discuss.pytorch.kr/">์ปค๋ฎค๋ํฐ</a>
</li>
</ul>
</div>
<a class="main-menu-open-button" href="#" data-behavior="open-mobile-menu"></a>
</div>
</div>
</div>
<body class="pytorch-body">
<div class="table-of-contents-link-wrapper">
<span>Table of Contents</span>
<a href="#" class="toggle-table-of-contents" data-behavior="toggle-table-of-contents"></a>
</div>
<nav data-toggle="wy-nav-shift" class="pytorch-left-menu" id="pytorch-left-menu">
<div class="pytorch-side-scroll">
<div class="pytorch-menu pytorch-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<div class="pytorch-left-menu-search">
<div class="version">
2.3.1+cu121
</div>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
<input type="text" name="q" placeholder="Search Tutorials" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<p class="caption" role="heading"><span class="caption-text">ํ์ดํ ์น(PyTorch) ๋ ์ํผ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../recipes/recipes_index.html">๋ชจ๋ ๋ ์ํผ ๋ณด๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../prototype/prototype_index.html">๋ชจ๋ ํ๋กํ ํ์
๋ ์ํผ ๋ณด๊ธฐ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">ํ์ดํ ์น(PyTorch) ์์ํ๊ธฐ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/intro.html">ํ์ดํ ์น(PyTorch) ๊ธฐ๋ณธ ์ตํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/quickstart_tutorial.html">๋น ๋ฅธ ์์(Quickstart)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/tensorqs_tutorial.html">ํ
์(Tensor)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/data_tutorial.html">Dataset๊ณผ DataLoader</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/transforms_tutorial.html">๋ณํ(Transform)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/buildmodel_tutorial.html">์ ๊ฒฝ๋ง ๋ชจ๋ธ ๊ตฌ์ฑํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/autogradqs_tutorial.html"><code class="docutils literal notranslate"><span class="pre">torch.autograd</span></code>๋ฅผ ์ฌ์ฉํ ์๋ ๋ฏธ๋ถ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/optimization_tutorial.html">๋ชจ๋ธ ๋งค๊ฐ๋ณ์ ์ต์ ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/basics/saveloadrun_tutorial.html">๋ชจ๋ธ ์ ์ฅํ๊ณ ๋ถ๋ฌ์ค๊ธฐ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Introduction to PyTorch on YouTube</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt.html">PyTorch ์๊ฐ - YouTube ์๋ฆฌ์ฆ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/introyt1_tutorial.html">PyTorch ์๊ฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/tensors_deeper_tutorial.html">Pytorch Tensor ์๊ฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/autogradyt_tutorial.html">The Fundamentals of Autograd</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/modelsyt_tutorial.html">Building Models with PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/tensorboardyt_tutorial.html">PyTorch TensorBoard Support</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/trainingyt.html">Training with PyTorch</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/introyt/captumyt.html">Model Understanding with Captum</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">ํ์ดํ ์น(PyTorch) ๋ฐฐ์ฐ๊ธฐ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deep_learning_60min_blitz.html">PyTorch๋ก ๋ฅ๋ฌ๋ํ๊ธฐ: 60๋ถ๋ง์ ๋์ฅ๋ด๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/pytorch_with_examples.html">์์ ๋ก ๋ฐฐ์ฐ๋ ํ์ดํ ์น(PyTorch)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/nn_tutorial.html"><cite>torch.nn</cite> ์ด <em>์ค์ ๋ก</em> ๋ฌด์์ธ๊ฐ์?</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard_tutorial.html">TensorBoard๋ก ๋ชจ๋ธ, ๋ฐ์ดํฐ, ํ์ต ์๊ฐํํ๊ธฐ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">์ด๋ฏธ์ง/๋น๋์ค</span></p>
<ul class="current">
<li class="toctree-l1 current"><a class="current reference internal" href="#">TorchVision Object Detection Finetuning Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/transfer_learning_tutorial.html">์ปดํจํฐ ๋น์ (Vision)์ ์ํ ์ ์ดํ์ต(Transfer Learning)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/fgsm_tutorial.html">์ ๋์ ์์ ์์ฑ(Adversarial Example Generation)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/dcgan_faces_tutorial.html">DCGAN ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/vt_tutorial.html">๋ฐฐํฌ๋ฅผ ์ํด ๋น์ ํธ๋์คํฌ๋จธ(Vision Transformer) ๋ชจ๋ธ ์ต์ ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="tiatoolbox_tutorial.html">Whole Slide Image Classification Using PyTorch and TIAToolbox</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">์ค๋์ค</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_io_tutorial.html">Audio I/O</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_resampling_tutorial.html">Audio Resampling</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_data_augmentation_tutorial.html">Audio Data Augmentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_feature_extractions_tutorial.html">Audio Feature Extractions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_feature_augmentation_tutorial.html">Audio Feature Augmentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/audio_datasets_tutorial.html">Audio Datasets</a></li>
<li class="toctree-l1"><a class="reference internal" href="speech_recognition_pipeline_tutorial.html">Speech Recognition with Wav2Vec2</a></li>
<li class="toctree-l1"><a class="reference internal" href="text_to_speech_with_torchaudio.html">Text-to-speech with Tacotron2</a></li>
<li class="toctree-l1"><a class="reference internal" href="forced_alignment_with_torchaudio_tutorial.html">wav2vec2์ ์ด์ฉํ ๊ฐ์ ์ ๋ ฌ</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">ํ
์คํธ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/bettertransformer_tutorial.html">Fast Transformer Inference with Better Transformer</a></li>
<li class="toctree-l1"><a class="reference internal" href="char_rnn_classification_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: ๋ฌธ์-๋จ์ RNN์ผ๋ก ์ด๋ฆ ๋ถ๋ฅํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="char_rnn_generation_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: ๋ฌธ์-๋จ์ RNN์ผ๋ก ์ด๋ฆ ์์ฑํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="seq2seq_translation_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: Sequence to Sequence ๋คํธ์ํฌ์ Attention์ ์ด์ฉํ ๋ฒ์ญ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/text_sentiment_ngrams_tutorial.html">torchtext ๋ผ์ด๋ธ๋ฌ๋ฆฌ๋ก ํ
์คํธ ๋ถ๋ฅํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/translation_transformer.html"><code class="docutils literal notranslate"><span class="pre">nn.Transformer</span></code> ์ torchtext๋ก ์ธ์ด ๋ฒ์ญํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/torchtext_custom_dataset_tutorial.html">Preprocess custom text dataset using Torchtext</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">๋ฐฑ์๋</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/onnx/intro_onnx.html">Introduction to ONNX</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">๊ฐํํ์ต</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="reinforcement_q_learning.html">๊ฐํ ํ์ต (DQN) ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="reinforcement_ppo.html">Reinforcement Learning (PPO) with TorchRL Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="mario_rl_tutorial.html">๋ง๋ฆฌ์ค ๊ฒ์ RL ์์ด์ ํธ๋ก ํ์ตํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/pendulum.html">Pendulum: Writing your environment and transforms with TorchRL</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">PyTorch ๋ชจ๋ธ์ ํ๋ก๋์
ํ๊ฒฝ์ ๋ฐฐํฌํ๊ธฐ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/onnx/intro_onnx.html">Introduction to ONNX</a></li>
<li class="toctree-l1"><a class="reference internal" href="flask_rest_api_tutorial.html">Flask๋ฅผ ์ฌ์ฉํ์ฌ Python์์ PyTorch๋ฅผ REST API๋ก ๋ฐฐํฌํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/Intro_to_TorchScript_tutorial.html">TorchScript ์๊ฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_export.html">C++์์ TorchScript ๋ชจ๋ธ ๋ก๋ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/super_resolution_with_onnxruntime.html">(์ ํ) PyTorch ๋ชจ๋ธ์ ONNX์ผ๋ก ๋ณํํ๊ณ ONNX ๋ฐํ์์์ ์คํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="realtime_rpi.html">Raspberry Pi 4 ์์ ์ค์๊ฐ ์ถ๋ก (Inference) (30fps!)</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">PyTorch ํ๋กํ์ผ๋ง</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/profiler.html">PyTorch ๋ชจ๋ ํ๋กํ์ผ๋งํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/hta_intro_tutorial.html">Introduction to Holistic Trace Analysis</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/hta_trace_diff_tutorial.html">Trace Diff using Holistic Trace Analysis</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Code Transforms with FX</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="fx_conv_bn_fuser.html">(๋ฒ ํ) FX์์ ํฉ์ฑ๊ณฑ/๋ฐฐ์น ์ ๊ทํ(Convolution/Batch Norm) ๊ฒฐํฉ๊ธฐ(Fuser) ๋ง๋ค๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="fx_profiling_tutorial.html">(beta) Building a Simple CPU Performance Profiler with FX</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">ํ๋ก ํธ์๋ API</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="memory_format_tutorial.html">(๋ฒ ํ) PyTorch๋ฅผ ์ฌ์ฉํ Channels Last ๋ฉ๋ชจ๋ฆฌ ํ์</a></li>
<li class="toctree-l1"><a class="reference internal" href="forward_ad_usage.html">Forward-mode Automatic Differentiation (Beta)</a></li>
<li class="toctree-l1"><a class="reference internal" href="jacobians_hessians.html">Jacobians, Hessians, hvp, vhp, and more: composing function transforms</a></li>
<li class="toctree-l1"><a class="reference internal" href="ensembling.html">๋ชจ๋ธ ์์๋ธ</a></li>
<li class="toctree-l1"><a class="reference internal" href="per_sample_grads.html">Per-sample-gradients</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_frontend.html">PyTorch C++ ํ๋ก ํธ์๋ ์ฌ์ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/torch-script-parallelism.html">TorchScript์ ๋์ ๋ณ๋ ฌ ์ฒ๋ฆฌ(Dynamic Parallelism)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_autograd.html">C++ ํ๋ก ํธ์๋์ ์๋ ๋ฏธ๋ถ (autograd)</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">PyTorch ํ์ฅํ๊ธฐ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="custom_function_double_backward_tutorial.html">Double Backward with Custom Functions</a></li>
<li class="toctree-l1"><a class="reference internal" href="custom_function_conv_bn_tutorial.html">Fusing Convolution and Batch Norm using Custom Function</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/cpp_extension.html">Custom C++ and CUDA Extensions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/torch_script_custom_ops.html">Extending TorchScript with Custom C++ Operators</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/torch_script_custom_classes.html">์ปค์คํ
C++ ํด๋์ค๋ก TorchScript ํ์ฅํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/dispatcher.html">Registering a Dispatched Operator in C++</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/extend_dispatcher.html">Extending dispatcher for a new backend in C++</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/privateuseone.html">Facilitating New Backend Integration by PrivateUse1</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">๋ชจ๋ธ ์ต์ ํ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/profiler.html">PyTorch ๋ชจ๋ ํ๋กํ์ผ๋งํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="tensorboard_profiler_tutorial.html">ํ
์๋ณด๋๋ฅผ ์ด์ฉํ ํ์ดํ ์น ํ๋กํ์ผ๋ฌ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/hyperparameter_tuning_tutorial.html">Ray Tune์ ์ฌ์ฉํ ํ์ดํผํ๋ผ๋ฏธํฐ ํ๋</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/vt_tutorial.html">๋ฐฐํฌ๋ฅผ ์ํด ๋น์ ํธ๋์คํฌ๋จธ(Vision Transformer) ๋ชจ๋ธ ์ต์ ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="parametrizations.html">Parametrizations Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="pruning_tutorial.html">๊ฐ์ง์น๊ธฐ ๊ธฐ๋ฒ(Pruning) ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/dynamic_quantization_tutorial.html">(๋ฒ ํ) LSTM ๊ธฐ๋ฐ ๋จ์ด ๋จ์ ์ธ์ด ๋ชจ๋ธ์ ๋์ ์์ํ</a></li>
<li class="toctree-l1"><a class="reference internal" href="dynamic_quantization_bert_tutorial.html">(๋ฒ ํ) BERT ๋ชจ๋ธ ๋์ ์์ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="quantized_transfer_learning_tutorial.html">(๋ฒ ํ) ์ปดํจํฐ ๋น์ ํํ ๋ฆฌ์ผ์ ์ํ ์์ํ๋ ์ ์ดํ์ต(Quantized Transfer Learning)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/static_quantization_tutorial.html">(๋ฒ ํ) PyTorch์์ Eager Mode๋ฅผ ์ด์ฉํ ์ ์ ์์ํ</a></li>
<li class="toctree-l1"><a class="reference internal" href="torchserve_with_ipex.html">Grokking PyTorch Intel CPU performance from first principles</a></li>
<li class="toctree-l1"><a class="reference internal" href="torchserve_with_ipex_2.html">Grokking PyTorch Intel CPU performance from first principles (Part 2)</a></li>
<li class="toctree-l1"><a class="reference internal" href="nvfuser_intro_tutorial.html">Getting Started - Accelerate Your Scripts with nvFuser</a></li>
<li class="toctree-l1"><a class="reference internal" href="ax_multiobjective_nas_tutorial.html">Multi-Objective NAS with Ax</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_compile_tutorial.html">Introduction to <code class="docutils literal notranslate"><span class="pre">torch.compile</span></code></a></li>
<li class="toctree-l1"><a class="reference internal" href="inductor_debug_cpu.html">Inductor CPU backend debugging and profiling</a></li>
<li class="toctree-l1"><a class="reference internal" href="scaled_dot_product_attention_tutorial.html">(Beta) Scaled Dot Product Attention (SDPA)๋ก ๊ณ ์ฑ๋ฅ ํธ๋์คํฌ๋จธ(Transformers) ๊ตฌํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="scaled_dot_product_attention_tutorial.html#torch-compile-sdpa"><code class="docutils literal notranslate"><span class="pre">torch.compile</span></code> ๊ณผ ํจ๊ป SDPA ์ฌ์ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="scaled_dot_product_attention_tutorial.html#sdpa-atteition-bias">SDPA๋ฅผ <code class="docutils literal notranslate"><span class="pre">atteition.bias</span></code> ํ์ ํด๋์ค์ ์ฌ์ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="scaled_dot_product_attention_tutorial.html#id8">๊ฒฐ๋ก </a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/knowledge_distillation_tutorial.html">Knowledge Distillation Tutorial</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">๋ณ๋ ฌ ๋ฐ ๋ถ์ฐ ํ์ต</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../distributed/home.html">Distributed and Parallel Training Tutorials</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/dist_overview.html">PyTorch Distributed Overview</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/ddp_series_intro.html">Distributed Data Parallel in PyTorch - Video Tutorials</a></li>
<li class="toctree-l1"><a class="reference internal" href="model_parallel_tutorial.html">๋จ์ผ ๋จธ์ ์ ์ฌ์ฉํ ๋ชจ๋ธ ๋ณ๋ ฌํ ๋ชจ๋ฒ ์ฌ๋ก</a></li>
<li class="toctree-l1"><a class="reference internal" href="ddp_tutorial.html">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ ์ฒ๋ฆฌ ์์ํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="dist_tuto.html">PyTorch๋ก ๋ถ์ฐ ์ดํ๋ฆฌ์ผ์ด์
๊ฐ๋ฐํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="FSDP_tutorial.html">Getting Started with Fully Sharded Data Parallel(FSDP)</a></li>
<li class="toctree-l1"><a class="reference internal" href="FSDP_adavnced_tutorial.html">Advanced Model Training with Fully Sharded Data Parallel (FSDP)</a></li>
<li class="toctree-l1"><a class="reference internal" href="TP_tutorial.html">Large Scale Transformer model training with Tensor Parallel (TP)</a></li>
<li class="toctree-l1"><a class="reference internal" href="process_group_cpp_extension_tutorial.html">Cpp ํ์ฅ์ ์ฌ์ฉํ ํ๋ก์ธ์ค ๊ทธ๋ฃน ๋ฐฑ์๋ ์ฌ์ฉ์ ์ ์</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_tutorial.html">Getting Started with Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_param_server_tutorial.html">Implementing a Parameter Server Using Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="dist_pipeline_parallel_tutorial.html">Distributed Pipeline Parallelism Using RPC</a></li>
<li class="toctree-l1"><a class="reference internal" href="rpc_async_execution.html">Implementing Batch RPC Processing Using Asynchronous Executions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/rpc_ddp_tutorial.html">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ(DDP)๊ณผ ๋ถ์ฐ RPC ํ๋ ์์ํฌ ๊ฒฐํฉ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/ddp_pipeline.html">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ ์ฒ๋ฆฌ์ ๋ณ๋ ฌ ์ฒ๋ฆฌ ํ์ดํ๋ผ์ธ์ ์ฌ์ฉํ ํธ๋์คํฌ๋จธ ๋ชจ๋ธ ํ์ต</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/generic_join.html">Distributed Training with Uneven Inputs Using the Join Context Manager</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Edge with ExecuTorch</span></p>
<ul>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/tutorials/export-to-executorch-tutorial.html">Exporting to ExecuTorch Tutorial</a></li>
<li class="toctree-l1"><a class="reference external" href=" https://pytorch.org/executorch/stable/running-a-model-cpp-tutorial.html">Running an ExecuTorch Model in C++ Tutorial</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/tutorials/sdk-integration-tutorial.html">Using the ExecuTorch SDK to Profile a Model</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/demo-apps-ios.html">Building an ExecuTorch iOS Demo App</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/demo-apps-android.html">Building an ExecuTorch Android Demo App</a></li>
<li class="toctree-l1"><a class="reference external" href="https://pytorch.org/executorch/stable/examples-end-to-end-to-lower-model-to-delegate.html">Lowering a Model as a Delegate</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">์ถ์ฒ ์์คํ
</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="torchrec_tutorial.html">TorchRec ์๊ฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../advanced/sharding.html">Exploring TorchRec sharding</a></li>
</ul>
<p class="caption" role="heading"><span class="caption-text">Multimodality</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/flava_finetuning_tutorial.html">TorchMultimodal ํํ ๋ฆฌ์ผ: FLAVA ๋ฏธ์ธ์กฐ์ </a></li>
</ul>
</div>
</div>
</nav>
<div class="pytorch-container">
<div class="pytorch-page-level-bar" id="pytorch-page-level-bar">
<div class="pytorch-breadcrumbs-wrapper">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="pytorch-breadcrumbs">
<li>
<a href="../index.html">
Tutorials
</a> >
</li>
<li>TorchVision Object Detection Finetuning Tutorial</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/intermediate/torchvision_tutorial.rst.txt" rel="nofollow"><img src="../_static/images/view-page-source-icon.svg"></a>
</li>
</ul>
</div>
</div>
<div class="pytorch-shortcuts-wrapper" id="pytorch-shortcuts-wrapper">
Shortcuts
</div>
</div>
<section data-toggle="wy-nav-shift" id="pytorch-content-wrap" class="pytorch-content-wrap">
<div class="pytorch-content-left">
<div class="pytorch-call-to-action-links">
<div id="tutorial-type">intermediate/torchvision_tutorial</div>
<div id="google-colab-link">
<img class="call-to-action-img" src="../_static/images/pytorch-colab.svg"/>
<div class="call-to-action-desktop-view">Run in Google Colab</div>
<div class="call-to-action-mobile-view">Colab</div>
</div>
<div id="download-notebook-link">
<img class="call-to-action-notebook-img" src="../_static/images/pytorch-download.svg"/>
<div class="call-to-action-desktop-view">Download Notebook</div>
<div class="call-to-action-mobile-view">Notebook</div>
</div>
<div id="github-view-link">
<img class="call-to-action-img" src="../_static/images/pytorch-github.svg"/>
<div class="call-to-action-desktop-view">View on GitHub</div>
<div class="call-to-action-mobile-view">GitHub</div>
</div>
</div>
<div class="rst-content">
<div role="main" class="main-content" itemscope="itemscope" itemtype="http://schema.org/Article">
<article itemprop="articleBody" id="pytorch-article" class="pytorch-article">
<div class="sphx-glr-download-link-note admonition note">
<p class="admonition-title">์ฐธ๊ณ </p>
<p>Click <a class="reference internal" href="#sphx-glr-download-intermediate-torchvision-tutorial-py"><span class="std std-ref">here</span></a>
to download the full example code</p>
</div>
<div class="sphx-glr-example-title section" id="torchvision-object-detection-finetuning-tutorial">
<span id="sphx-glr-intermediate-torchvision-tutorial-py"></span><h1>TorchVision Object Detection Finetuning Tutorial<a class="headerlink" href="#torchvision-object-detection-finetuning-tutorial" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h1>
<p>For this tutorial, we will be finetuning a pre-trained <a class="reference external" href="https://arxiv.org/abs/1703.06870">Mask
R-CNN</a> model on the <a class="reference external" href="https://www.cis.upenn.edu/~jshi/ped_html/">Penn-Fudan
Database for Pedestrian Detection and
Segmentation</a>. It contains
170 images with 345 instances of pedestrians, and we will use it to
illustrate how to use the new features in torchvision in order to train
an object detection and instance segmentation model on a custom dataset.</p>
<div class="admonition note">
<p class="admonition-title">์ฐธ๊ณ </p>
<p>This tutorial works only with torchvision version >=0.16 or nightly.
If youโre using torchvision<=0.15, please follow
<a class="reference external" href="https://github.com/pytorch/tutorials/blob/d686b662932a380a58b7683425faa00c06bcf502/intermediate_source/torchvision_tutorial.rst">this tutorial instead</a>.</p>
</div>
<div class="section" id="defining-the-dataset">
<h2>Defining the Dataset<a class="headerlink" href="#defining-the-dataset" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>The reference scripts for training object detection, instance
segmentation and person keypoint detection allows for easily supporting
adding new custom datasets. The dataset should inherit from the standard
<a class="reference external" href="https://pytorch.org/docs/stable/data.html#torch.utils.data.Dataset" title="(PyTorch v2.3์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.utils.data.Dataset</span></code></a> class, and implement <code class="docutils literal notranslate"><span class="pre">__len__</span></code> and
<code class="docutils literal notranslate"><span class="pre">__getitem__</span></code>.</p>
<p>The only specificity that we require is that the dataset <code class="docutils literal notranslate"><span class="pre">__getitem__</span></code>
should return a tuple:</p>
<ul class="simple">
<li><p>image: <a class="reference external" href="https://pytorch.org/vision/stable/generated/torchvision.tv_tensors.Image.html#torchvision.tv_tensors.Image" title="(Torchvision v0.18์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torchvision.tv_tensors.Image</span></code></a> of shape <code class="docutils literal notranslate"><span class="pre">[3,</span> <span class="pre">H,</span> <span class="pre">W]</span></code>, a pure tensor, or a PIL Image of size <code class="docutils literal notranslate"><span class="pre">(H,</span> <span class="pre">W)</span></code></p></li>
<li><p>target: a dict containing the following fields</p>
<ul>
<li><p><code class="docutils literal notranslate"><span class="pre">boxes</span></code>, <a class="reference external" href="https://pytorch.org/vision/stable/generated/torchvision.tv_tensors.BoundingBoxes.html#torchvision.tv_tensors.BoundingBoxes" title="(Torchvision v0.18์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torchvision.tv_tensors.BoundingBoxes</span></code></a> of shape <code class="docutils literal notranslate"><span class="pre">[N,</span> <span class="pre">4]</span></code>:
the coordinates of the <code class="docutils literal notranslate"><span class="pre">N</span></code> bounding boxes in <code class="docutils literal notranslate"><span class="pre">[x0,</span> <span class="pre">y0,</span> <span class="pre">x1,</span> <span class="pre">y1]</span></code> format, ranging from <code class="docutils literal notranslate"><span class="pre">0</span></code>
to <code class="docutils literal notranslate"><span class="pre">W</span></code> and <code class="docutils literal notranslate"><span class="pre">0</span></code> to <code class="docutils literal notranslate"><span class="pre">H</span></code></p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">labels</span></code>, integer <a class="reference external" href="https://pytorch.org/docs/stable/tensors.html#torch.Tensor" title="(PyTorch v2.3์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.Tensor</span></code></a> of shape <code class="docutils literal notranslate"><span class="pre">[N]</span></code>: the label for each bounding box.
<code class="docutils literal notranslate"><span class="pre">0</span></code> represents always the background class.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">image_id</span></code>, int: an image identifier. It should be
unique between all the images in the dataset, and is used during
evaluation</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">area</span></code>, float <a class="reference external" href="https://pytorch.org/docs/stable/tensors.html#torch.Tensor" title="(PyTorch v2.3์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.Tensor</span></code></a> of shape <code class="docutils literal notranslate"><span class="pre">[N]</span></code>: the area of the bounding box. This is used
during evaluation with the COCO metric, to separate the metric
scores between small, medium and large boxes.</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">iscrowd</span></code>, uint8 <a class="reference external" href="https://pytorch.org/docs/stable/tensors.html#torch.Tensor" title="(PyTorch v2.3์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.Tensor</span></code></a> of shape <code class="docutils literal notranslate"><span class="pre">[N]</span></code>: instances with <code class="docutils literal notranslate"><span class="pre">iscrowd=True</span></code> will be
ignored during evaluation.</p></li>
<li><p>(optionally) <code class="docutils literal notranslate"><span class="pre">masks</span></code>, <a class="reference external" href="https://pytorch.org/vision/stable/generated/torchvision.tv_tensors.Mask.html#torchvision.tv_tensors.Mask" title="(Torchvision v0.18์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torchvision.tv_tensors.Mask</span></code></a> of shape <code class="docutils literal notranslate"><span class="pre">[N,</span> <span class="pre">H,</span> <span class="pre">W]</span></code>: the segmentation
masks for each one of the objects</p></li>
</ul>
</li>
</ul>
<p>If your dataset is compliant with above requirements then it will work for both
training and evaluation codes from the reference script. Evaluation code will use scripts from
<code class="docutils literal notranslate"><span class="pre">pycocotools</span></code> which can be installed with <code class="docutils literal notranslate"><span class="pre">pip</span> <span class="pre">install</span> <span class="pre">pycocotools</span></code>.</p>
<div class="admonition note">
<p class="admonition-title">์ฐธ๊ณ </p>
<p>For Windows, please install <code class="docutils literal notranslate"><span class="pre">pycocotools</span></code> from <a class="reference external" href="https://github.com/gautamchitnis/cocoapi">gautamchitnis</a> with command</p>
<p><code class="docutils literal notranslate"><span class="pre">pip</span> <span class="pre">install</span> <span class="pre">git+https://github.com/gautamchitnis/cocoapi.git@cocodataset-master#subdirectory=PythonAPI</span></code></p>
</div>
<p>One note on the <code class="docutils literal notranslate"><span class="pre">labels</span></code>. The model considers class <code class="docutils literal notranslate"><span class="pre">0</span></code> as background. If your dataset does not contain the background class,
you should not have <code class="docutils literal notranslate"><span class="pre">0</span></code> in your <code class="docutils literal notranslate"><span class="pre">labels</span></code>. For example, assuming you have just two classes, <em>cat</em> and <em>dog</em>, you can
define <code class="docutils literal notranslate"><span class="pre">1</span></code> (not <code class="docutils literal notranslate"><span class="pre">0</span></code>) to represent <em>cats</em> and <code class="docutils literal notranslate"><span class="pre">2</span></code> to represent <em>dogs</em>. So, for instance, if one of the images has both
classes, your <code class="docutils literal notranslate"><span class="pre">labels</span></code> tensor should look like <code class="docutils literal notranslate"><span class="pre">[1,</span> <span class="pre">2]</span></code>.</p>
<p>Additionally, if you want to use aspect ratio grouping during training
(so that each batch only contains images with similar aspect ratios),
then it is recommended to also implement a <code class="docutils literal notranslate"><span class="pre">get_height_and_width</span></code>
method, which returns the height and the width of the image. If this
method is not provided, we query all elements of the dataset via
<code class="docutils literal notranslate"><span class="pre">__getitem__</span></code> , which loads the image in memory and is slower than if
a custom method is provided.</p>
<div class="section" id="writing-a-custom-dataset-for-pennfudan">
<h3>Writing a custom dataset for PennFudan<a class="headerlink" href="#writing-a-custom-dataset-for-pennfudan" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>Letโs write a dataset for the PennFudan dataset. First, letโs download the dataset and
extract the <a class="reference external" href="https://www.cis.upenn.edu/~jshi/ped_html/PennFudanPed.zip">zip file</a>:</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">wget</span> <span class="n">https</span><span class="p">:</span><span class="o">//</span><span class="n">www</span><span class="o">.</span><span class="n">cis</span><span class="o">.</span><span class="n">upenn</span><span class="o">.</span><span class="n">edu</span><span class="o">/~</span><span class="n">jshi</span><span class="o">/</span><span class="n">ped_html</span><span class="o">/</span><span class="n">PennFudanPed</span><span class="o">.</span><span class="n">zip</span> <span class="o">-</span><span class="n">P</span> <span class="n">data</span>
<span class="n">cd</span> <span class="n">data</span> <span class="o">&&</span> <span class="n">unzip</span> <span class="n">PennFudanPed</span><span class="o">.</span><span class="n">zip</span>
</pre></div>
</div>
<p>We have the following folder structure:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">PennFudanPed</span><span class="o">/</span>
<span class="n">PedMasks</span><span class="o">/</span>
<span class="n">FudanPed00001_mask</span><span class="o">.</span><span class="n">png</span>
<span class="n">FudanPed00002_mask</span><span class="o">.</span><span class="n">png</span>
<span class="n">FudanPed00003_mask</span><span class="o">.</span><span class="n">png</span>
<span class="n">FudanPed00004_mask</span><span class="o">.</span><span class="n">png</span>
<span class="o">...</span>
<span class="n">PNGImages</span><span class="o">/</span>
<span class="n">FudanPed00001</span><span class="o">.</span><span class="n">png</span>
<span class="n">FudanPed00002</span><span class="o">.</span><span class="n">png</span>
<span class="n">FudanPed00003</span><span class="o">.</span><span class="n">png</span>
<span class="n">FudanPed00004</span><span class="o">.</span><span class="n">png</span>
</pre></div>
</div>
<p>Here is one example of a pair of images and segmentation masks</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">matplotlib.pyplot</span> <span class="k">as</span> <span class="nn">plt</span>
<span class="kn">from</span> <span class="nn">torchvision.io</span> <span class="kn">import</span> <span class="n">read_image</span>
<span class="n">image</span> <span class="o">=</span> <span class="n">read_image</span><span class="p">(</span><span class="s2">"data/PennFudanPed/PNGImages/FudanPed00046.png"</span><span class="p">)</span>
<span class="n">mask</span> <span class="o">=</span> <span class="n">read_image</span><span class="p">(</span><span class="s2">"data/PennFudanPed/PedMasks/FudanPed00046_mask.png"</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">figure</span><span class="p">(</span><span class="n">figsize</span><span class="o">=</span><span class="p">(</span><span class="mi">16</span><span class="p">,</span> <span class="mi">8</span><span class="p">))</span>
<span class="n">plt</span><span class="o">.</span><span class="n">subplot</span><span class="p">(</span><span class="mi">121</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">title</span><span class="p">(</span><span class="s2">"Image"</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">imshow</span><span class="p">(</span><span class="n">image</span><span class="o">.</span><span class="n">permute</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">0</span><span class="p">))</span>
<span class="n">plt</span><span class="o">.</span><span class="n">subplot</span><span class="p">(</span><span class="mi">122</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">title</span><span class="p">(</span><span class="s2">"Mask"</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">imshow</span><span class="p">(</span><span class="n">mask</span><span class="o">.</span><span class="n">permute</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">0</span><span class="p">))</span>
</pre></div>
</div>
<p>So each image has a corresponding
segmentation mask, where each color correspond to a different instance.
Letโs write a <a class="reference external" href="https://pytorch.org/docs/stable/data.html#torch.utils.data.Dataset" title="(PyTorch v2.3์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.utils.data.Dataset</span></code></a> class for this dataset.
In the code below, we are wrapping images, bounding boxes and masks into
<a class="reference external" href="https://pytorch.org/vision/stable/generated/torchvision.tv_tensors.TVTensor.html#torchvision.tv_tensors.TVTensor" title="(Torchvision v0.18์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torchvision.tv_tensors.TVTensor</span></code></a> classes so that we will be able to apply torchvision
built-in transformations (<a class="reference external" href="https://pytorch.org/vision/stable/transforms.html">new Transforms API</a>)
for the given object detection and segmentation task.
Namely, image tensors will be wrapped by <a class="reference external" href="https://pytorch.org/vision/stable/generated/torchvision.tv_tensors.Image.html#torchvision.tv_tensors.Image" title="(Torchvision v0.18์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torchvision.tv_tensors.Image</span></code></a>, bounding boxes into
<a class="reference external" href="https://pytorch.org/vision/stable/generated/torchvision.tv_tensors.BoundingBoxes.html#torchvision.tv_tensors.BoundingBoxes" title="(Torchvision v0.18์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torchvision.tv_tensors.BoundingBoxes</span></code></a> and masks into <a class="reference external" href="https://pytorch.org/vision/stable/generated/torchvision.tv_tensors.Mask.html#torchvision.tv_tensors.Mask" title="(Torchvision v0.18์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torchvision.tv_tensors.Mask</span></code></a>.
As <a class="reference external" href="https://pytorch.org/vision/stable/generated/torchvision.tv_tensors.TVTensor.html#torchvision.tv_tensors.TVTensor" title="(Torchvision v0.18์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torchvision.tv_tensors.TVTensor</span></code></a> are <a class="reference external" href="https://pytorch.org/docs/stable/tensors.html#torch.Tensor" title="(PyTorch v2.3์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.Tensor</span></code></a> subclasses, wrapped objects are also tensors and inherit the plain
<a class="reference external" href="https://pytorch.org/docs/stable/tensors.html#torch.Tensor" title="(PyTorch v2.3์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.Tensor</span></code></a> API. For more information about torchvision <code class="docutils literal notranslate"><span class="pre">tv_tensors</span></code> see
<a class="reference external" href="https://pytorch.org/vision/main/auto_examples/transforms/plot_transforms_getting_started.html#what-are-tvtensors">this documentation</a>.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">os</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">from</span> <span class="nn">torchvision.io</span> <span class="kn">import</span> <span class="n">read_image</span>
<span class="kn">from</span> <span class="nn">torchvision.ops.boxes</span> <span class="kn">import</span> <span class="n">masks_to_boxes</span>
<span class="kn">from</span> <span class="nn">torchvision</span> <span class="kn">import</span> <span class="n">tv_tensors</span>
<span class="kn">from</span> <span class="nn">torchvision.transforms.v2</span> <span class="kn">import</span> <span class="n">functional</span> <span class="k">as</span> <span class="n">F</span>
<span class="k">class</span> <span class="nc">PennFudanDataset</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">Dataset</span><span class="p">):</span>
<span class="k">def</span> <span class="fm">__init__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">root</span><span class="p">,</span> <span class="n">transforms</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">root</span> <span class="o">=</span> <span class="n">root</span>
<span class="bp">self</span><span class="o">.</span><span class="n">transforms</span> <span class="o">=</span> <span class="n">transforms</span>
<span class="c1"># load all image files, sorting them to</span>
<span class="c1"># ensure that they are aligned</span>
<span class="bp">self</span><span class="o">.</span><span class="n">imgs</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">sorted</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">listdir</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">root</span><span class="p">,</span> <span class="s2">"PNGImages"</span><span class="p">))))</span>
<span class="bp">self</span><span class="o">.</span><span class="n">masks</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="nb">sorted</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">listdir</span><span class="p">(</span><span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="n">root</span><span class="p">,</span> <span class="s2">"PedMasks"</span><span class="p">))))</span>
<span class="k">def</span> <span class="fm">__getitem__</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">idx</span><span class="p">):</span>
<span class="c1"># load images and masks</span>
<span class="n">img_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">root</span><span class="p">,</span> <span class="s2">"PNGImages"</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">imgs</span><span class="p">[</span><span class="n">idx</span><span class="p">])</span>
<span class="n">mask_path</span> <span class="o">=</span> <span class="n">os</span><span class="o">.</span><span class="n">path</span><span class="o">.</span><span class="n">join</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">root</span><span class="p">,</span> <span class="s2">"PedMasks"</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">masks</span><span class="p">[</span><span class="n">idx</span><span class="p">])</span>
<span class="n">img</span> <span class="o">=</span> <span class="n">read_image</span><span class="p">(</span><span class="n">img_path</span><span class="p">)</span>
<span class="n">mask</span> <span class="o">=</span> <span class="n">read_image</span><span class="p">(</span><span class="n">mask_path</span><span class="p">)</span>
<span class="c1"># instances are encoded as different colors</span>
<span class="n">obj_ids</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">unique</span><span class="p">(</span><span class="n">mask</span><span class="p">)</span>
<span class="c1"># first id is the background, so remove it</span>
<span class="n">obj_ids</span> <span class="o">=</span> <span class="n">obj_ids</span><span class="p">[</span><span class="mi">1</span><span class="p">:]</span>
<span class="n">num_objs</span> <span class="o">=</span> <span class="nb">len</span><span class="p">(</span><span class="n">obj_ids</span><span class="p">)</span>
<span class="c1"># split the color-encoded mask into a set</span>
<span class="c1"># of binary masks</span>
<span class="n">masks</span> <span class="o">=</span> <span class="p">(</span><span class="n">mask</span> <span class="o">==</span> <span class="n">obj_ids</span><span class="p">[:,</span> <span class="kc">None</span><span class="p">,</span> <span class="kc">None</span><span class="p">])</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span>
<span class="c1"># get bounding box coordinates for each mask</span>
<span class="n">boxes</span> <span class="o">=</span> <span class="n">masks_to_boxes</span><span class="p">(</span><span class="n">masks</span><span class="p">)</span>
<span class="c1"># there is only one class</span>
<span class="n">labels</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">ones</span><span class="p">((</span><span class="n">num_objs</span><span class="p">,),</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">int64</span><span class="p">)</span>
<span class="n">image_id</span> <span class="o">=</span> <span class="n">idx</span>
<span class="n">area</span> <span class="o">=</span> <span class="p">(</span><span class="n">boxes</span><span class="p">[:,</span> <span class="mi">3</span><span class="p">]</span> <span class="o">-</span> <span class="n">boxes</span><span class="p">[:,</span> <span class="mi">1</span><span class="p">])</span> <span class="o">*</span> <span class="p">(</span><span class="n">boxes</span><span class="p">[:,</span> <span class="mi">2</span><span class="p">]</span> <span class="o">-</span> <span class="n">boxes</span><span class="p">[:,</span> <span class="mi">0</span><span class="p">])</span>
<span class="c1"># suppose all instances are not crowd</span>
<span class="n">iscrowd</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">zeros</span><span class="p">((</span><span class="n">num_objs</span><span class="p">,),</span> <span class="n">dtype</span><span class="o">=</span><span class="n">torch</span><span class="o">.</span><span class="n">int64</span><span class="p">)</span>
<span class="c1"># Wrap sample and targets into torchvision tv_tensors:</span>
<span class="n">img</span> <span class="o">=</span> <span class="n">tv_tensors</span><span class="o">.</span><span class="n">Image</span><span class="p">(</span><span class="n">img</span><span class="p">)</span>
<span class="n">target</span> <span class="o">=</span> <span class="p">{}</span>
<span class="n">target</span><span class="p">[</span><span class="s2">"boxes"</span><span class="p">]</span> <span class="o">=</span> <span class="n">tv_tensors</span><span class="o">.</span><span class="n">BoundingBoxes</span><span class="p">(</span><span class="n">boxes</span><span class="p">,</span> <span class="nb">format</span><span class="o">=</span><span class="s2">"XYXY"</span><span class="p">,</span> <span class="n">canvas_size</span><span class="o">=</span><span class="n">F</span><span class="o">.</span><span class="n">get_size</span><span class="p">(</span><span class="n">img</span><span class="p">))</span>
<span class="n">target</span><span class="p">[</span><span class="s2">"masks"</span><span class="p">]</span> <span class="o">=</span> <span class="n">tv_tensors</span><span class="o">.</span><span class="n">Mask</span><span class="p">(</span><span class="n">masks</span><span class="p">)</span>
<span class="n">target</span><span class="p">[</span><span class="s2">"labels"</span><span class="p">]</span> <span class="o">=</span> <span class="n">labels</span>
<span class="n">target</span><span class="p">[</span><span class="s2">"image_id"</span><span class="p">]</span> <span class="o">=</span> <span class="n">image_id</span>
<span class="n">target</span><span class="p">[</span><span class="s2">"area"</span><span class="p">]</span> <span class="o">=</span> <span class="n">area</span>
<span class="n">target</span><span class="p">[</span><span class="s2">"iscrowd"</span><span class="p">]</span> <span class="o">=</span> <span class="n">iscrowd</span>
<span class="k">if</span> <span class="bp">self</span><span class="o">.</span><span class="n">transforms</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</span><span class="p">:</span>
<span class="n">img</span><span class="p">,</span> <span class="n">target</span> <span class="o">=</span> <span class="bp">self</span><span class="o">.</span><span class="n">transforms</span><span class="p">(</span><span class="n">img</span><span class="p">,</span> <span class="n">target</span><span class="p">)</span>
<span class="k">return</span> <span class="n">img</span><span class="p">,</span> <span class="n">target</span>
<span class="k">def</span> <span class="fm">__len__</span><span class="p">(</span><span class="bp">self</span><span class="p">):</span>
<span class="k">return</span> <span class="nb">len</span><span class="p">(</span><span class="bp">self</span><span class="o">.</span><span class="n">imgs</span><span class="p">)</span>
</pre></div>
</div>
<p>Thatโs all for the dataset. Now letโs define a model that can perform
predictions on this dataset.</p>
</div>
</div>
<div class="section" id="defining-your-model">
<h2>Defining your model<a class="headerlink" href="#defining-your-model" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>In this tutorial, we will be using <a class="reference external" href="https://arxiv.org/abs/1703.06870">Mask
R-CNN</a>, which is based on top of
<a class="reference external" href="https://arxiv.org/abs/1506.01497">Faster R-CNN</a>. Faster R-CNN is a
model that predicts both bounding boxes and class scores for potential
objects in the image.</p>
<img alt="../_static/img/tv_tutorial/tv_image03.png" src="../_static/img/tv_tutorial/tv_image03.png" />
<p>Mask R-CNN adds an extra branch
into Faster R-CNN, which also predicts segmentation masks for each
instance.</p>
<img alt="../_static/img/tv_tutorial/tv_image04.png" src="../_static/img/tv_tutorial/tv_image04.png" />
<p>There are two common
situations where one might want
to modify one of the available models in TorchVision Model Zoo. The first
is when we want to start from a pre-trained model, and just finetune the
last layer. The other is when we want to replace the backbone of the
model with a different one (for faster predictions, for example).</p>
<p>Letโs go see how we would do one or another in the following sections.</p>
<div class="section" id="finetuning-from-a-pretrained-model">
<h3>1 - Finetuning from a pretrained model<a class="headerlink" href="#finetuning-from-a-pretrained-model" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>Letโs suppose that you want to start from a model pre-trained on COCO
and want to finetune it for your particular classes. Here is a possible
way of doing it:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torchvision</span>
<span class="kn">from</span> <span class="nn">torchvision.models.detection.faster_rcnn</span> <span class="kn">import</span> <span class="n">FastRCNNPredictor</span>
<span class="c1"># load a model pre-trained on COCO</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">torchvision</span><span class="o">.</span><span class="n">models</span><span class="o">.</span><span class="n">detection</span><span class="o">.</span><span class="n">fasterrcnn_resnet50_fpn</span><span class="p">(</span><span class="n">weights</span><span class="o">=</span><span class="s2">"DEFAULT"</span><span class="p">)</span>
<span class="c1"># replace the classifier with a new one, that has</span>
<span class="c1"># num_classes which is user-defined</span>
<span class="n">num_classes</span> <span class="o">=</span> <span class="mi">2</span> <span class="c1"># 1 class (person) + background</span>
<span class="c1"># get number of input features for the classifier</span>
<span class="n">in_features</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">roi_heads</span><span class="o">.</span><span class="n">box_predictor</span><span class="o">.</span><span class="n">cls_score</span><span class="o">.</span><span class="n">in_features</span>
<span class="c1"># replace the pre-trained head with a new one</span>
<span class="n">model</span><span class="o">.</span><span class="n">roi_heads</span><span class="o">.</span><span class="n">box_predictor</span> <span class="o">=</span> <span class="n">FastRCNNPredictor</span><span class="p">(</span><span class="n">in_features</span><span class="p">,</span> <span class="n">num_classes</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="modifying-the-model-to-add-a-different-backbone">
<h3>2 - Modifying the model to add a different backbone<a class="headerlink" href="#modifying-the-model-to-add-a-different-backbone" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torchvision</span>
<span class="kn">from</span> <span class="nn">torchvision.models.detection</span> <span class="kn">import</span> <span class="n">FasterRCNN</span>
<span class="kn">from</span> <span class="nn">torchvision.models.detection.rpn</span> <span class="kn">import</span> <span class="n">AnchorGenerator</span>
<span class="c1"># load a pre-trained model for classification and return</span>
<span class="c1"># only the features</span>
<span class="n">backbone</span> <span class="o">=</span> <span class="n">torchvision</span><span class="o">.</span><span class="n">models</span><span class="o">.</span><span class="n">mobilenet_v2</span><span class="p">(</span><span class="n">weights</span><span class="o">=</span><span class="s2">"DEFAULT"</span><span class="p">)</span><span class="o">.</span><span class="n">features</span>
<span class="c1"># ``FasterRCNN`` needs to know the number of</span>
<span class="c1"># output channels in a backbone. For mobilenet_v2, it's 1280</span>
<span class="c1"># so we need to add it here</span>
<span class="n">backbone</span><span class="o">.</span><span class="n">out_channels</span> <span class="o">=</span> <span class="mi">1280</span>
<span class="c1"># let's make the RPN generate 5 x 3 anchors per spatial</span>
<span class="c1"># location, with 5 different sizes and 3 different aspect</span>
<span class="c1"># ratios. We have a Tuple[Tuple[int]] because each feature</span>
<span class="c1"># map could potentially have different sizes and</span>
<span class="c1"># aspect ratios</span>
<span class="n">anchor_generator</span> <span class="o">=</span> <span class="n">AnchorGenerator</span><span class="p">(</span>
<span class="n">sizes</span><span class="o">=</span><span class="p">((</span><span class="mi">32</span><span class="p">,</span> <span class="mi">64</span><span class="p">,</span> <span class="mi">128</span><span class="p">,</span> <span class="mi">256</span><span class="p">,</span> <span class="mi">512</span><span class="p">),),</span>
<span class="n">aspect_ratios</span><span class="o">=</span><span class="p">((</span><span class="mf">0.5</span><span class="p">,</span> <span class="mf">1.0</span><span class="p">,</span> <span class="mf">2.0</span><span class="p">),)</span>
<span class="p">)</span>
<span class="c1"># let's define what are the feature maps that we will</span>
<span class="c1"># use to perform the region of interest cropping, as well as</span>
<span class="c1"># the size of the crop after rescaling.</span>
<span class="c1"># if your backbone returns a Tensor, featmap_names is expected to</span>
<span class="c1"># be [0]. More generally, the backbone should return an</span>
<span class="c1"># ``OrderedDict[Tensor]``, and in ``featmap_names`` you can choose which</span>
<span class="c1"># feature maps to use.</span>
<span class="n">roi_pooler</span> <span class="o">=</span> <span class="n">torchvision</span><span class="o">.</span><span class="n">ops</span><span class="o">.</span><span class="n">MultiScaleRoIAlign</span><span class="p">(</span>
<span class="n">featmap_names</span><span class="o">=</span><span class="p">[</span><span class="s1">'0'</span><span class="p">],</span>
<span class="n">output_size</span><span class="o">=</span><span class="mi">7</span><span class="p">,</span>
<span class="n">sampling_ratio</span><span class="o">=</span><span class="mi">2</span>
<span class="p">)</span>
<span class="c1"># put the pieces together inside a Faster-RCNN model</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">FasterRCNN</span><span class="p">(</span>
<span class="n">backbone</span><span class="p">,</span>
<span class="n">num_classes</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span>
<span class="n">rpn_anchor_generator</span><span class="o">=</span><span class="n">anchor_generator</span><span class="p">,</span>
<span class="n">box_roi_pool</span><span class="o">=</span><span class="n">roi_pooler</span>
<span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="object-detection-and-instance-segmentation-model-for-pennfudan-dataset">
<h3>Object detection and instance segmentation model for PennFudan Dataset<a class="headerlink" href="#object-detection-and-instance-segmentation-model-for-pennfudan-dataset" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>In our case, we want to finetune from a pre-trained model, given that
our dataset is very small, so we will be following approach number 1.</p>
<p>Here we want to also compute the instance segmentation masks, so we will
be using Mask R-CNN:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">torchvision</span>
<span class="kn">from</span> <span class="nn">torchvision.models.detection.faster_rcnn</span> <span class="kn">import</span> <span class="n">FastRCNNPredictor</span>
<span class="kn">from</span> <span class="nn">torchvision.models.detection.mask_rcnn</span> <span class="kn">import</span> <span class="n">MaskRCNNPredictor</span>
<span class="k">def</span> <span class="nf">get_model_instance_segmentation</span><span class="p">(</span><span class="n">num_classes</span><span class="p">):</span>
<span class="c1"># load an instance segmentation model pre-trained on COCO</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">torchvision</span><span class="o">.</span><span class="n">models</span><span class="o">.</span><span class="n">detection</span><span class="o">.</span><span class="n">maskrcnn_resnet50_fpn</span><span class="p">(</span><span class="n">weights</span><span class="o">=</span><span class="s2">"DEFAULT"</span><span class="p">)</span>
<span class="c1"># get number of input features for the classifier</span>
<span class="n">in_features</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">roi_heads</span><span class="o">.</span><span class="n">box_predictor</span><span class="o">.</span><span class="n">cls_score</span><span class="o">.</span><span class="n">in_features</span>
<span class="c1"># replace the pre-trained head with a new one</span>
<span class="n">model</span><span class="o">.</span><span class="n">roi_heads</span><span class="o">.</span><span class="n">box_predictor</span> <span class="o">=</span> <span class="n">FastRCNNPredictor</span><span class="p">(</span><span class="n">in_features</span><span class="p">,</span> <span class="n">num_classes</span><span class="p">)</span>
<span class="c1"># now get the number of input features for the mask classifier</span>
<span class="n">in_features_mask</span> <span class="o">=</span> <span class="n">model</span><span class="o">.</span><span class="n">roi_heads</span><span class="o">.</span><span class="n">mask_predictor</span><span class="o">.</span><span class="n">conv5_mask</span><span class="o">.</span><span class="n">in_channels</span>
<span class="n">hidden_layer</span> <span class="o">=</span> <span class="mi">256</span>
<span class="c1"># and replace the mask predictor with a new one</span>
<span class="n">model</span><span class="o">.</span><span class="n">roi_heads</span><span class="o">.</span><span class="n">mask_predictor</span> <span class="o">=</span> <span class="n">MaskRCNNPredictor</span><span class="p">(</span>
<span class="n">in_features_mask</span><span class="p">,</span>
<span class="n">hidden_layer</span><span class="p">,</span>
<span class="n">num_classes</span>
<span class="p">)</span>
<span class="k">return</span> <span class="n">model</span>
</pre></div>
</div>
<p>Thatโs it, this will make <code class="docutils literal notranslate"><span class="pre">model</span></code> be ready to be trained and evaluated
on your custom dataset.</p>
</div>
</div>
<div class="section" id="putting-everything-together">
<h2>Putting everything together<a class="headerlink" href="#putting-everything-together" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>In <code class="docutils literal notranslate"><span class="pre">references/detection/</span></code>, we have a number of helper functions to
simplify training and evaluating detection models. Here, we will use
<code class="docutils literal notranslate"><span class="pre">references/detection/engine.py</span></code> and <code class="docutils literal notranslate"><span class="pre">references/detection/utils.py</span></code>.
Just download everything under <code class="docutils literal notranslate"><span class="pre">references/detection</span></code> to your folder and use them here.
On Linux if you have <code class="docutils literal notranslate"><span class="pre">wget</span></code>, you can download them using below commands:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">os</span><span class="o">.</span><span class="n">system</span><span class="p">(</span><span class="s2">"wget https://raw.githubusercontent.com/pytorch/vision/main/references/detection/engine.py"</span><span class="p">)</span>
<span class="n">os</span><span class="o">.</span><span class="n">system</span><span class="p">(</span><span class="s2">"wget https://raw.githubusercontent.com/pytorch/vision/main/references/detection/utils.py"</span><span class="p">)</span>
<span class="n">os</span><span class="o">.</span><span class="n">system</span><span class="p">(</span><span class="s2">"wget https://raw.githubusercontent.com/pytorch/vision/main/references/detection/coco_utils.py"</span><span class="p">)</span>
<span class="n">os</span><span class="o">.</span><span class="n">system</span><span class="p">(</span><span class="s2">"wget https://raw.githubusercontent.com/pytorch/vision/main/references/detection/coco_eval.py"</span><span class="p">)</span>
<span class="n">os</span><span class="o">.</span><span class="n">system</span><span class="p">(</span><span class="s2">"wget https://raw.githubusercontent.com/pytorch/vision/main/references/detection/transforms.py"</span><span class="p">)</span>
</pre></div>
</div>
<p>Since v0.15.0 torchvision provides <a class="reference external" href="https://pytorch.org/vision/stable/transforms.html">new Transforms API</a>
to easily write data augmentation pipelines for Object Detection and Segmentation tasks.</p>
<p>Letโs write some helper functions for data augmentation /
transformation:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">torchvision.transforms</span> <span class="kn">import</span> <span class="n">v2</span> <span class="k">as</span> <span class="n">T</span>
<span class="k">def</span> <span class="nf">get_transform</span><span class="p">(</span><span class="n">train</span><span class="p">):</span>
<span class="n">transforms</span> <span class="o">=</span> <span class="p">[]</span>
<span class="k">if</span> <span class="n">train</span><span class="p">:</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">T</span><span class="o">.</span><span class="n">RandomHorizontalFlip</span><span class="p">(</span><span class="mf">0.5</span><span class="p">))</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">T</span><span class="o">.</span><span class="n">ToDtype</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">float</span><span class="p">,</span> <span class="n">scale</span><span class="o">=</span><span class="kc">True</span><span class="p">))</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">T</span><span class="o">.</span><span class="n">ToPureTensor</span><span class="p">())</span>
<span class="k">return</span> <span class="n">T</span><span class="o">.</span><span class="n">Compose</span><span class="p">(</span><span class="n">transforms</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="testing-forward-method-optional">
<h2>Testing <code class="docutils literal notranslate"><span class="pre">forward()</span></code> method (Optional)<a class="headerlink" href="#testing-forward-method-optional" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>Before iterating over the dataset, itโs good to see what the model
expects during training and inference time on sample data.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">utils</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">torchvision</span><span class="o">.</span><span class="n">models</span><span class="o">.</span><span class="n">detection</span><span class="o">.</span><span class="n">fasterrcnn_resnet50_fpn</span><span class="p">(</span><span class="n">weights</span><span class="o">=</span><span class="s2">"DEFAULT"</span><span class="p">)</span>
<span class="n">dataset</span> <span class="o">=</span> <span class="n">PennFudanDataset</span><span class="p">(</span><span class="s1">'data/PennFudanPed'</span><span class="p">,</span> <span class="n">get_transform</span><span class="p">(</span><span class="n">train</span><span class="o">=</span><span class="kc">True</span><span class="p">))</span>
<span class="n">data_loader</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">DataLoader</span><span class="p">(</span>
<span class="n">dataset</span><span class="p">,</span>
<span class="n">batch_size</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span>
<span class="n">shuffle</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">num_workers</span><span class="o">=</span><span class="mi">4</span><span class="p">,</span>
<span class="n">collate_fn</span><span class="o">=</span><span class="n">utils</span><span class="o">.</span><span class="n">collate_fn</span>
<span class="p">)</span>
<span class="c1"># For Training</span>
<span class="n">images</span><span class="p">,</span> <span class="n">targets</span> <span class="o">=</span> <span class="nb">next</span><span class="p">(</span><span class="nb">iter</span><span class="p">(</span><span class="n">data_loader</span><span class="p">))</span>
<span class="n">images</span> <span class="o">=</span> <span class="nb">list</span><span class="p">(</span><span class="n">image</span> <span class="k">for</span> <span class="n">image</span> <span class="ow">in</span> <span class="n">images</span><span class="p">)</span>
<span class="n">targets</span> <span class="o">=</span> <span class="p">[{</span><span class="n">k</span><span class="p">:</span> <span class="n">v</span> <span class="k">for</span> <span class="n">k</span><span class="p">,</span> <span class="n">v</span> <span class="ow">in</span> <span class="n">t</span><span class="o">.</span><span class="n">items</span><span class="p">()}</span> <span class="k">for</span> <span class="n">t</span> <span class="ow">in</span> <span class="n">targets</span><span class="p">]</span>
<span class="n">output</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="n">images</span><span class="p">,</span> <span class="n">targets</span><span class="p">)</span> <span class="c1"># Returns losses and detections</span>
<span class="nb">print</span><span class="p">(</span><span class="n">output</span><span class="p">)</span>
<span class="c1"># For inference</span>
<span class="n">model</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="n">x</span> <span class="o">=</span> <span class="p">[</span><span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="mi">300</span><span class="p">,</span> <span class="mi">400</span><span class="p">),</span> <span class="n">torch</span><span class="o">.</span><span class="n">rand</span><span class="p">(</span><span class="mi">3</span><span class="p">,</span> <span class="mi">500</span><span class="p">,</span> <span class="mi">400</span><span class="p">)]</span>
<span class="n">predictions</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="n">x</span><span class="p">)</span> <span class="c1"># Returns predictions</span>
<span class="nb">print</span><span class="p">(</span><span class="n">predictions</span><span class="p">[</span><span class="mi">0</span><span class="p">])</span>
</pre></div>
</div>
<p>Letโs now write the main function which performs the training and the
validation:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">engine</span> <span class="kn">import</span> <span class="n">train_one_epoch</span><span class="p">,</span> <span class="n">evaluate</span>
<span class="c1"># train on the GPU or on the CPU, if a GPU is not available</span>
<span class="n">device</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cuda'</span><span class="p">)</span> <span class="k">if</span> <span class="n">torch</span><span class="o">.</span><span class="n">cuda</span><span class="o">.</span><span class="n">is_available</span><span class="p">()</span> <span class="k">else</span> <span class="n">torch</span><span class="o">.</span><span class="n">device</span><span class="p">(</span><span class="s1">'cpu'</span><span class="p">)</span>
<span class="c1"># our dataset has two classes only - background and person</span>
<span class="n">num_classes</span> <span class="o">=</span> <span class="mi">2</span>
<span class="c1"># use our dataset and defined transformations</span>
<span class="n">dataset</span> <span class="o">=</span> <span class="n">PennFudanDataset</span><span class="p">(</span><span class="s1">'data/PennFudanPed'</span><span class="p">,</span> <span class="n">get_transform</span><span class="p">(</span><span class="n">train</span><span class="o">=</span><span class="kc">True</span><span class="p">))</span>
<span class="n">dataset_test</span> <span class="o">=</span> <span class="n">PennFudanDataset</span><span class="p">(</span><span class="s1">'data/PennFudanPed'</span><span class="p">,</span> <span class="n">get_transform</span><span class="p">(</span><span class="n">train</span><span class="o">=</span><span class="kc">False</span><span class="p">))</span>
<span class="c1"># split the dataset in train and test set</span>
<span class="n">indices</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">randperm</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">dataset</span><span class="p">))</span><span class="o">.</span><span class="n">tolist</span><span class="p">()</span>
<span class="n">dataset</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">Subset</span><span class="p">(</span><span class="n">dataset</span><span class="p">,</span> <span class="n">indices</span><span class="p">[:</span><span class="o">-</span><span class="mi">50</span><span class="p">])</span>
<span class="n">dataset_test</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">Subset</span><span class="p">(</span><span class="n">dataset_test</span><span class="p">,</span> <span class="n">indices</span><span class="p">[</span><span class="o">-</span><span class="mi">50</span><span class="p">:])</span>
<span class="c1"># define training and validation data loaders</span>
<span class="n">data_loader</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">DataLoader</span><span class="p">(</span>
<span class="n">dataset</span><span class="p">,</span>
<span class="n">batch_size</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span>
<span class="n">shuffle</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
<span class="n">num_workers</span><span class="o">=</span><span class="mi">4</span><span class="p">,</span>
<span class="n">collate_fn</span><span class="o">=</span><span class="n">utils</span><span class="o">.</span><span class="n">collate_fn</span>
<span class="p">)</span>
<span class="n">data_loader_test</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">utils</span><span class="o">.</span><span class="n">data</span><span class="o">.</span><span class="n">DataLoader</span><span class="p">(</span>
<span class="n">dataset_test</span><span class="p">,</span>
<span class="n">batch_size</span><span class="o">=</span><span class="mi">1</span><span class="p">,</span>
<span class="n">shuffle</span><span class="o">=</span><span class="kc">False</span><span class="p">,</span>
<span class="n">num_workers</span><span class="o">=</span><span class="mi">4</span><span class="p">,</span>
<span class="n">collate_fn</span><span class="o">=</span><span class="n">utils</span><span class="o">.</span><span class="n">collate_fn</span>
<span class="p">)</span>
<span class="c1"># get the model using our helper function</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">get_model_instance_segmentation</span><span class="p">(</span><span class="n">num_classes</span><span class="p">)</span>
<span class="c1"># move model to the right device</span>
<span class="n">model</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</span><span class="p">)</span>
<span class="c1"># construct an optimizer</span>
<span class="n">params</span> <span class="o">=</span> <span class="p">[</span><span class="n">p</span> <span class="k">for</span> <span class="n">p</span> <span class="ow">in</span> <span class="n">model</span><span class="o">.</span><span class="n">parameters</span><span class="p">()</span> <span class="k">if</span> <span class="n">p</span><span class="o">.</span><span class="n">requires_grad</span><span class="p">]</span>
<span class="n">optimizer</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">optim</span><span class="o">.</span><span class="n">SGD</span><span class="p">(</span>
<span class="n">params</span><span class="p">,</span>
<span class="n">lr</span><span class="o">=</span><span class="mf">0.005</span><span class="p">,</span>
<span class="n">momentum</span><span class="o">=</span><span class="mf">0.9</span><span class="p">,</span>
<span class="n">weight_decay</span><span class="o">=</span><span class="mf">0.0005</span>
<span class="p">)</span>
<span class="c1"># and a learning rate scheduler</span>
<span class="n">lr_scheduler</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">optim</span><span class="o">.</span><span class="n">lr_scheduler</span><span class="o">.</span><span class="n">StepLR</span><span class="p">(</span>
<span class="n">optimizer</span><span class="p">,</span>
<span class="n">step_size</span><span class="o">=</span><span class="mi">3</span><span class="p">,</span>
<span class="n">gamma</span><span class="o">=</span><span class="mf">0.1</span>
<span class="p">)</span>
<span class="c1"># let's train it just for 2 epochs</span>
<span class="n">num_epochs</span> <span class="o">=</span> <span class="mi">2</span>
<span class="k">for</span> <span class="n">epoch</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="n">num_epochs</span><span class="p">):</span>
<span class="c1"># train for one epoch, printing every 10 iterations</span>
<span class="n">train_one_epoch</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">optimizer</span><span class="p">,</span> <span class="n">data_loader</span><span class="p">,</span> <span class="n">device</span><span class="p">,</span> <span class="n">epoch</span><span class="p">,</span> <span class="n">print_freq</span><span class="o">=</span><span class="mi">10</span><span class="p">)</span>
<span class="c1"># update the learning rate</span>
<span class="n">lr_scheduler</span><span class="o">.</span><span class="n">step</span><span class="p">()</span>
<span class="c1"># evaluate on the test dataset</span>
<span class="n">evaluate</span><span class="p">(</span><span class="n">model</span><span class="p">,</span> <span class="n">data_loader_test</span><span class="p">,</span> <span class="n">device</span><span class="o">=</span><span class="n">device</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">"That's it!"</span><span class="p">)</span>
</pre></div>
</div>
<p>So after one epoch of training, we obtain a COCO-style mAP > 50, and
a mask mAP of 65.</p>
<p>But what do the predictions look like? Letโs take one image in the
dataset and verify</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">import</span> <span class="nn">matplotlib.pyplot</span> <span class="k">as</span> <span class="nn">plt</span>
<span class="kn">from</span> <span class="nn">torchvision.utils</span> <span class="kn">import</span> <span class="n">draw_bounding_boxes</span><span class="p">,</span> <span class="n">draw_segmentation_masks</span>
<span class="n">image</span> <span class="o">=</span> <span class="n">read_image</span><span class="p">(</span><span class="s2">"data/PennFudanPed/PNGImages/FudanPed00046.png"</span><span class="p">)</span>
<span class="n">eval_transform</span> <span class="o">=</span> <span class="n">get_transform</span><span class="p">(</span><span class="n">train</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="n">model</span><span class="o">.</span><span class="n">eval</span><span class="p">()</span>
<span class="k">with</span> <span class="n">torch</span><span class="o">.</span><span class="n">no_grad</span><span class="p">():</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">eval_transform</span><span class="p">(</span><span class="n">image</span><span class="p">)</span>
<span class="c1"># convert RGBA -> RGB and move to device</span>
<span class="n">x</span> <span class="o">=</span> <span class="n">x</span><span class="p">[:</span><span class="mi">3</span><span class="p">,</span> <span class="o">...</span><span class="p">]</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">device</span><span class="p">)</span>
<span class="n">predictions</span> <span class="o">=</span> <span class="n">model</span><span class="p">([</span><span class="n">x</span><span class="p">,</span> <span class="p">])</span>
<span class="n">pred</span> <span class="o">=</span> <span class="n">predictions</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="n">image</span> <span class="o">=</span> <span class="p">(</span><span class="mf">255.0</span> <span class="o">*</span> <span class="p">(</span><span class="n">image</span> <span class="o">-</span> <span class="n">image</span><span class="o">.</span><span class="n">min</span><span class="p">())</span> <span class="o">/</span> <span class="p">(</span><span class="n">image</span><span class="o">.</span><span class="n">max</span><span class="p">()</span> <span class="o">-</span> <span class="n">image</span><span class="o">.</span><span class="n">min</span><span class="p">()))</span><span class="o">.</span><span class="n">to</span><span class="p">(</span><span class="n">torch</span><span class="o">.</span><span class="n">uint8</span><span class="p">)</span>
<span class="n">image</span> <span class="o">=</span> <span class="n">image</span><span class="p">[:</span><span class="mi">3</span><span class="p">,</span> <span class="o">...</span><span class="p">]</span>
<span class="n">pred_labels</span> <span class="o">=</span> <span class="p">[</span><span class="sa">f</span><span class="s2">"pedestrian: </span><span class="si">{</span><span class="n">score</span><span class="si">:</span><span class="s2">.3f</span><span class="si">}</span><span class="s2">"</span> <span class="k">for</span> <span class="n">label</span><span class="p">,</span> <span class="n">score</span> <span class="ow">in</span> <span class="nb">zip</span><span class="p">(</span><span class="n">pred</span><span class="p">[</span><span class="s2">"labels"</span><span class="p">],</span> <span class="n">pred</span><span class="p">[</span><span class="s2">"scores"</span><span class="p">])]</span>
<span class="n">pred_boxes</span> <span class="o">=</span> <span class="n">pred</span><span class="p">[</span><span class="s2">"boxes"</span><span class="p">]</span><span class="o">.</span><span class="n">long</span><span class="p">()</span>
<span class="n">output_image</span> <span class="o">=</span> <span class="n">draw_bounding_boxes</span><span class="p">(</span><span class="n">image</span><span class="p">,</span> <span class="n">pred_boxes</span><span class="p">,</span> <span class="n">pred_labels</span><span class="p">,</span> <span class="n">colors</span><span class="o">=</span><span class="s2">"red"</span><span class="p">)</span>
<span class="n">masks</span> <span class="o">=</span> <span class="p">(</span><span class="n">pred</span><span class="p">[</span><span class="s2">"masks"</span><span class="p">]</span> <span class="o">></span> <span class="mf">0.7</span><span class="p">)</span><span class="o">.</span><span class="n">squeeze</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
<span class="n">output_image</span> <span class="o">=</span> <span class="n">draw_segmentation_masks</span><span class="p">(</span><span class="n">output_image</span><span class="p">,</span> <span class="n">masks</span><span class="p">,</span> <span class="n">alpha</span><span class="o">=</span><span class="mf">0.5</span><span class="p">,</span> <span class="n">colors</span><span class="o">=</span><span class="s2">"blue"</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">figure</span><span class="p">(</span><span class="n">figsize</span><span class="o">=</span><span class="p">(</span><span class="mi">12</span><span class="p">,</span> <span class="mi">12</span><span class="p">))</span>
<span class="n">plt</span><span class="o">.</span><span class="n">imshow</span><span class="p">(</span><span class="n">output_image</span><span class="o">.</span><span class="n">permute</span><span class="p">(</span><span class="mi">1</span><span class="p">,</span> <span class="mi">2</span><span class="p">,</span> <span class="mi">0</span><span class="p">))</span>
</pre></div>
</div>
<p>The results look good!</p>
</div>
<div class="section" id="wrapping-up">
<h2>Wrapping up<a class="headerlink" href="#wrapping-up" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>In this tutorial, you have learned how to create your own training
pipeline for object detection models on a custom dataset. For
that, you wrote a <a class="reference external" href="https://pytorch.org/docs/stable/data.html#torch.utils.data.Dataset" title="(PyTorch v2.3์์)"><code class="xref py py-class docutils literal notranslate"><span class="pre">torch.utils.data.Dataset</span></code></a> class that returns the
images and the ground truth boxes and segmentation masks. You also
leveraged a Mask R-CNN model pre-trained on COCO train2017 in order to
perform transfer learning on this new dataset.</p>
<p>For a more complete example, which includes multi-machine / multi-GPU
training, check <code class="docutils literal notranslate"><span class="pre">references/detection/train.py</span></code>, which is present in
the torchvision repository.</p>
<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 0 minutes 0.000 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-intermediate-torchvision-tutorial-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../_downloads/7590258df9f28b5ae0994c3b5b035edf/torchvision_tutorial.py"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Python</span> <span class="pre">source</span> <span class="pre">code:</span> <span class="pre">torchvision_tutorial.py</span></code></a></p>
</div>
<div class="sphx-glr-download sphx-glr-download-jupyter docutils container">
<p><a class="reference download internal" download="" href="../_downloads/4a542c9f39bedbfe7de5061767181d36/torchvision_tutorial.ipynb"><code class="xref download docutils literal notranslate"><span class="pre">Download</span> <span class="pre">Jupyter</span> <span class="pre">notebook:</span> <span class="pre">torchvision_tutorial.ipynb</span></code></a></p>
</div>
</div>
<p class="sphx-glr-signature"><a class="reference external" href="https://sphinx-gallery.github.io">Gallery generated by Sphinx-Gallery</a></p>
</div>
</div>
</article>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="../beginner/transfer_learning_tutorial.html" class="btn btn-neutral float-right" title="์ปดํจํฐ ๋น์ (Vision)์ ์ํ ์ ์ดํ์ต(Transfer Learning)" accesskey="n" rel="next">Next <img src="../_static/images/chevron-right-orange.svg" class="next-page"></a>
<a href="tensorboard_tutorial.html" class="btn btn-neutral" title="TensorBoard๋ก ๋ชจ๋ธ, ๋ฐ์ดํฐ, ํ์ต ์๊ฐํํ๊ธฐ" accesskey="p" rel="prev"><img src="../_static/images/chevron-right-orange.svg" class="previous-page"> Previous</a>
</div>
<hr class="community-hr hr-top" />
<div class="community-container">
<div class="community-prompt">๋ ๊ถ๊ธํ์๊ฑฐ๋ ๊ฐ์ ํ ๋ด์ฉ์ด ์์ผ์ ๊ฐ์? ์ปค๋ฎค๋ํฐ์ ์ฐธ์ฌํด๋ณด์ธ์!</div>
<div class="community-link"><a href="https://discuss.pytorch.kr/" aria-label="PyTorchKoreaCommunity">ํ๊ตญ์ด ์ปค๋ฎค๋ํฐ ๋ฐ๋ก๊ฐ๊ธฐ</a></div>
</div>
<hr class="community-hr hr-bottom"/>
<hr class="rating-hr hr-top" />
<div class="rating-container">
<div class="rating-prompt">์ด ํํ ๋ฆฌ์ผ์ด ์ด๋ ์
จ๋์? ํ๊ฐํด์ฃผ์๋ฉด ์ดํ ๊ฐ์ ์ ์ฐธ๊ณ ํ๊ฒ ์ต๋๋ค! :)</div>
<div class="stars-outer">
<i class="far fa-star" title="1 Star" data-behavior="tutorial-rating" data-count="1"></i>
<i class="far fa-star" title="2 Stars" data-behavior="tutorial-rating" data-count="2"></i>
<i class="far fa-star" title="3 Stars" data-behavior="tutorial-rating" data-count="3"></i>
<i class="far fa-star" title="4 Stars" data-behavior="tutorial-rating" data-count="4"></i>
<i class="far fa-star" title="5 Stars" data-behavior="tutorial-rating" data-count="5"></i>
</div>
</div>
<hr class="rating-hr hr-bottom"/>
<div role="contentinfo">
<p>
© Copyright 2018-2024, PyTorch & ํ์ดํ ์น ํ๊ตญ ์ฌ์ฉ์ ๋ชจ์(PyTorch Korea User Group).
</p>
</div>
<div>
Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>.
</div>
</footer>
</div>
<script>