-
Notifications
You must be signed in to change notification settings - Fork 257
/
Copy pathneural_style_tutorial.html
1261 lines (1033 loc) ยท 101 KB
/
neural_style_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="PyTorch๋ฅผ ์ด์ฉํ ๋ด๋ด ๋ณํ(Neural Transfer)" />
<meta property="og:type" content="article" />
<meta property="og:url" content="https://tutorials.pytorch.kr/advanced/neural_style_tutorial.html" />
<meta property="og:description" content="Author: Alexis Jacq Edited by: Winston Herring ๋ฒ์ญ: ์ ์ฌ๋ฏผ ์๊ฐ: ์ด๋ฒ ํํ ๋ฆฌ์ผ์ Leon A. Gatys, Alexander S. Ecker and Matthias Bethge์ ์ํด ๊ฐ๋ฐ๋ ๋ด๋ด ์คํ์ผ(Neural-Style) ์๊ณ ๋ฆฌ์ฆ ์ ๊ตฌํํ๋ ๋ฐฉ๋ฒ์ ๋ํ์ฌ ์ค๋ช
ํฉ๋๋ค. ๋ด๋ด ์คํ์ผ(Neural-Style), ๋๋ ๋ด๋ด ๋ณํ(Neural-Transfer)์ ์ฌ์ฉํ๋ฉด ์ด๋ฏธ์ง๋ฅผ ์๋ก์ด ์์ ์ ์คํ์ผ๋ก ์ฌํํ ์ ์์ต๋๋ค. ์ด ์๊ณ ๋ฆฌ์ฆ์ ์
๋ ฅ ์ด๋ฏธ์ง, Content ์ด๋ฏธ์ง, Style ..." />
<meta property="og:image" content="https://tutorials.pytorch.kr/_static/logos/logo-kr-sm-dark.png" />
<meta property="og:image:alt" content="PyTorch๋ฅผ ์ด์ฉํ ๋ด๋ด ๋ณํ(Neural Transfer)" />
<meta property="og:ignore_canonical" content="true" />
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>PyTorch๋ฅผ ์ด์ฉํ ๋ด๋ด ๋ณํ(Neural Transfer) — ํ์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@0.13.11/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="index" title="์์ธ" href="../genindex.html" />
<link rel="search" title="๊ฒ์" href="../search.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">
2.0.1
</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="../intermediate/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>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/torchvision_tutorial.html">TorchVision ๊ฐ์ฒด ๊ฒ์ถ ๋ฏธ์ธ์กฐ์ (Finetuning) ํํ ๋ฆฌ์ผ</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 ์ํ ๋น์ ํธ๋์คํฌ๋จธ(Vision Transformer) ๋ชจ๋ธ ์ต์ ํํ๊ธฐ</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="../intermediate/speech_recognition_pipeline_tutorial.html">Speech Recognition with Wav2Vec2</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/text_to_speech_with_torchaudio.html">Text-to-speech with Tacotron2</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/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/transformer_tutorial.html"><code class="docutils literal notranslate"><span class="pre">nn.Transformer</span></code> ์ torchtext๋ก ์ํ์ค-ํฌ-์ํ์ค(Sequence-to-Sequence) ๋ชจ๋ธ๋งํ๊ธฐ</a></li>
<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="../intermediate/char_rnn_classification_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: ๋ฌธ์-๋จ์ RNN์ผ๋ก ์ด๋ฆ ๋ถ๋ฅํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/char_rnn_generation_tutorial.html">๊ธฐ์ด๋ถํฐ ์์ํ๋ NLP: ๋ฌธ์-๋จ์ RNN์ผ๋ก ์ด๋ฆ ์์ฑํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/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>
</ul>
<p class="caption" role="heading"><span class="caption-text">๊ฐํํ์ต</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/reinforcement_q_learning.html">๊ฐํ ํ์ต (DQN) ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/reinforcement_ppo.html">Reinforcement Learning (PPO) with TorchRL Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/mario_rl_tutorial.html">๋ง๋ฆฌ์ค ๊ฒ์ RL ์์ด์ ํธ๋ก ํ์ตํ๊ธฐ</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="../intermediate/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="cpp_export.html">C++์์ TorchScript ๋ชจ๋ธ ๋ก๋ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="super_resolution_with_onnxruntime.html">(์ ํ) PyTorch ๋ชจ๋ธ์ ONNX์ผ๋ก ๋ณํํ๊ณ ONNX ๋ฐํ์์์ ์คํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/realtime_rpi.html">Raspberry Pi 4 ์์ ์ค์๊ฐ ์ถ๋ก (Inference) (30fps!)</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="../intermediate/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="../intermediate/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="../intermediate/memory_format_tutorial.html">(๋ฒ ํ) PyTorch๋ฅผ ์ฌ์ฉํ Channels Last ๋ฉ๋ชจ๋ฆฌ ํ์</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/forward_ad_usage.html">Forward-mode Automatic Differentiation (Beta)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/jacobians_hessians.html">Jacobians, Hessians, hvp, vhp, and more: composing function transforms</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/ensembling.html">๋ชจ๋ธ ์์๋ธ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/per_sample_grads.html">Per-sample-gradients</a></li>
<li class="toctree-l1"><a class="reference internal" href="cpp_frontend.html">PyTorch C++ ํ๋ก ํธ์๋ ์ฌ์ฉํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch-script-parallelism.html">TorchScript์ ๋์ ๋ณ๋ ฌ ์ฒ๋ฆฌ(Dynamic Parallelism)</a></li>
<li class="toctree-l1"><a class="reference internal" href="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="../intermediate/custom_function_double_backward_tutorial.html">Double Backward with Custom Functions</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/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="cpp_extension.html">Custom C++ and CUDA Extensions</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_script_custom_ops.html">Extending TorchScript with Custom C++ Operators</a></li>
<li class="toctree-l1"><a class="reference internal" href="torch_script_custom_classes.html">์ปค์คํ
C++ ํด๋์ค๋ก TorchScript ํ์ฅํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="dispatcher.html">Registering a Dispatched Operator in C++</a></li>
<li class="toctree-l1"><a class="reference internal" href="extend_dispatcher.html">Extending dispatcher for a new backend in C++</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 ํ๋กํ์ผ๋ง ํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/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 ์ด์ฉํ ํ์ดํผํ๋ผ๋ฏธํฐ ํ๋</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/vt_tutorial.html">๋ฐฐํฌ๋ฅผ ์ํ ๋น์ ํธ๋์คํฌ๋จธ(Vision Transformer) ๋ชจ๋ธ ์ต์ ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/parametrizations.html">Parametrizations Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/pruning_tutorial.html">๊ฐ์ง์น๊ธฐ ๊ธฐ๋ฒ(Pruning) ํํ ๋ฆฌ์ผ</a></li>
<li class="toctree-l1"><a class="reference internal" href="dynamic_quantization_tutorial.html">(๋ฒ ํ) LSTM ๊ธฐ๋ฐ ๋จ์ด ๋จ์ ์ธ์ด ๋ชจ๋ธ์ ๋์ ์์ํ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dynamic_quantization_bert_tutorial.html">(๋ฒ ํ) BERT ๋ชจ๋ธ ๋์ ์์ํํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/quantized_transfer_learning_tutorial.html">(๋ฒ ํ) ์ปดํจํฐ ๋น์ ํํ ๋ฆฌ์ผ์ ์ํ ์์ํ๋ ์ ์ดํ์ต(Quantized Transfer Learning)</a></li>
<li class="toctree-l1"><a class="reference internal" href="static_quantization_tutorial.html">(๋ฒ ํ) PyTorch์์ Eager Mode๋ฅผ ์ด์ฉํ ์ ์ ์์ํ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/torchserve_with_ipex.html">Grokking PyTorch Intel CPU performance from first principles</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/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="../intermediate/nvfuser_intro_tutorial.html">Getting Started - Accelerate Your Scripts with nvFuser</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/ax_multiobjective_nas_tutorial.html">Multi-Objective NAS with Ax</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/torch_compile_tutorial.html">torch.compile Tutorial</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/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="../intermediate/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="../intermediate/scaled_dot_product_attention_tutorial.html#id6">๊ฒฐ๋ก </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="../intermediate/model_parallel_tutorial.html">๋จ์ผ ๋จธ์ ์ ์ฌ์ฉํ ๋ชจ๋ธ ๋ณ๋ ฌํ ๋ชจ๋ฒ ์ฌ๋ก</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/ddp_tutorial.html">๋ถ์ฐ ๋ฐ์ดํฐ ๋ณ๋ ฌ ์ฒ๋ฆฌ ์์ํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dist_tuto.html">PyTorch๋ก ๋ถ์ฐ ์ดํ๋ฆฌ์ผ์ด์
๊ฐ๋ฐํ๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/FSDP_tutorial.html">Getting Started with Fully Sharded Data Parallel(FSDP)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/FSDP_adavnced_tutorial.html">Advanced Model Training with Fully Sharded Data Parallel (FSDP)</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/process_group_cpp_extension_tutorial.html">Cpp ํ์ฅ์ ์ฌ์ฉํ์ฌ ํ๋ก์ธ์ค ๊ทธ๋ฃน ๋ฐฑ์๋ ์ฌ์ฉ์ ์ ์</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/rpc_tutorial.html">Getting Started with Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/rpc_param_server_tutorial.html">Implementing a Parameter Server Using Distributed RPC Framework</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/dist_pipeline_parallel_tutorial.html">Distributed Pipeline Parallelism Using RPC</a></li>
<li class="toctree-l1"><a class="reference internal" href="../intermediate/rpc_async_execution.html">Implementing Batch RPC Processing Using Asynchronous Executions</a></li>
<li class="toctree-l1"><a class="reference internal" href="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="../intermediate/pipeline_tutorial.html">ํ์ดํ๋ผ์ธ ๋ณ๋ ฌํ๋ก ํธ๋์คํฌ๋จธ ๋ชจ๋ธ ํ์ต์ํค๊ธฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="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="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">๋ชจ๋ฐ์ผ</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deeplabv3_on_ios.html">iOS์์์ ์ด๋ฏธ์ง ๋ถํ DeepLapV3</a></li>
<li class="toctree-l1"><a class="reference internal" href="../beginner/deeplabv3_on_android.html">์๋๋ก์ด๋์์์ ์ด๋ฏธ์ง ๋ถํ DeepLapV3</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="../intermediate/torchrec_tutorial.html">TorchRec ์๊ฐ</a></li>
<li class="toctree-l1"><a class="reference internal" href="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>PyTorch๋ฅผ ์ด์ฉํ ๋ด๋ด ๋ณํ(Neural Transfer)</li>
<li class="pytorch-breadcrumbs-aside">
<a href="../_sources/advanced/neural_style_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">advanced/neural_style_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-advanced-neural-style-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="pytorch-neural-transfer">
<span id="sphx-glr-advanced-neural-style-tutorial-py"></span><h1>PyTorch๋ฅผ ์ด์ฉํ ๋ด๋ด ๋ณํ(Neural Transfer)<a class="headerlink" href="#pytorch-neural-transfer" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h1>
<p><strong>Author</strong>: <a class="reference external" href="https://alexis-jacq.github.io">Alexis Jacq</a></p>
<p><strong>Edited by</strong>: <a class="reference external" href="https://github.com/winston6">Winston Herring</a></p>
<p><strong>๋ฒ์ญ</strong>: <a class="reference external" href="https://github.com/jjeamin">์ ์ฌ๋ฏผ</a></p>
<div class="section" id="id2">
<h2>์๊ฐ<a class="headerlink" href="#id2" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>์ด๋ฒ ํํ ๋ฆฌ์ผ์ Leon A. Gatys, Alexander S. Ecker and Matthias Bethge์ ์ํด ๊ฐ๋ฐ๋
<a class="reference external" href="https://arxiv.org/abs/1508.06576">๋ด๋ด ์คํ์ผ(Neural-Style) ์๊ณ ๋ฆฌ์ฆ</a> ์ ๊ตฌํํ๋ ๋ฐฉ๋ฒ์ ๋ํ์ฌ ์ค๋ช
ํฉ๋๋ค.
๋ด๋ด ์คํ์ผ(Neural-Style), ๋๋ ๋ด๋ด ๋ณํ(Neural-Transfer)์ ์ฌ์ฉํ๋ฉด ์ด๋ฏธ์ง๋ฅผ ์๋ก์ด ์์ ์ ์คํ์ผ๋ก ์ฌํํ ์ ์์ต๋๋ค.
์ด ์๊ณ ๋ฆฌ์ฆ์ ์
๋ ฅ ์ด๋ฏธ์ง, Content ์ด๋ฏธ์ง, Style ์ด๋ฏธ์ง 3๊ฐ์ ์ด๋ฏธ์ง๋ฅผ ๊ฐ์ ธ์์ Content ์ด๋ฏธ์ง์ Content์ Style ์ด๋ฏธ์ง์ ์์ ์ Style์ ๋ฎ๋๋ก ์
๋ ฅ์ ๋ณํํฉ๋๋ค.</p>
<div class="figure align-default">
<img alt="content1" src="../_images/neuralstyle.png" />
</div>
</div>
<div class="section" id="id3">
<h2>๊ธฐ๋ณธ ์๋ฆฌ<a class="headerlink" href="#id3" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>์๋ฆฌ๋ ๊ฐ๋จํฉ๋๋ค.
Content(<span class="math">\(D_C\)</span>)์ Style(<span class="math">\(D_S\)</span>)์ ๋ํด ํ๋์ฉ ๋ ๊ฐ์ ๊ฑฐ๋ฆฌ๋ฅผ ์ ์ํฉ๋๋ค.
<span class="math">\(D_C\)</span> ๋ ๋ ์ด๋ฏธ์ง ๊ฐ์ Content์ ์ฐจ์ด๋ฅผ ์ธก์ ํ๊ณ <span class="math">\(D_S\)</span> ๋ ๋ ์ด๋ฏธ์ง ๊ฐ์ Style์ ์ฐจ์ด๋ฅผ ์ธก์ ํฉ๋๋ค.
๊ทธ๋ฐ ๋ค์, ์
๋ ฅ์ผ๋ก ์ฌ์ฉ๋๋ ์ธ ๋ฒ์งธ ์ด๋ฏธ์ง๋ฅผ ๊ฐ์ ธ์์
Content ์ด๋ฏธ์ง๋ก Content ๊ฑฐ๋ฆฌ์ Style ์ด๋ฏธ์ง๋ก Style ๊ฑฐ๋ฆฌ๋ฅผ ์ต์ํํ๋๋ก ๋ณํํฉ๋๋ค.
์ด์ ํ์ํ ํจํค์ง๋ฅผ ๊ฐ์ ธ์์ ๋ด๋ด ๋ณํ(neural transfer)์ ์์ํ ์ ์์ต๋๋ค.</p>
</div>
<div class="section" id="id4">
<h2>ํจํค์ง ๋ถ๋ฌ์ค๊ธฐ ๋ฐ ์ฅ์น ์ ํ<a class="headerlink" href="#id4" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>๋ค์์ ๋ด๋ด ๋ณํ(neural transfer)์ ๊ตฌํํ๋๋ฐ ํ์ํ ํจํค์ง ๋ชฉ๋ก์
๋๋ค.</p>
<ul class="simple">
<li><p><code class="docutils literal notranslate"><span class="pre">torch</span></code>, <code class="docutils literal notranslate"><span class="pre">torch.nn</span></code>, <code class="docutils literal notranslate"><span class="pre">numpy</span></code> (PyTorch๋ก ์ ๊ฒฝ๋ง์ ๊ตฌํํ๊ธฐ ์ํ ํ์ ํจํค์ง)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torch.optim</span></code> (ํจ์จ์ ์ธ ๊ฒฝ์ฌ ํ๊ฐ๋ฒ)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">PIL</span></code>, <code class="docutils literal notranslate"><span class="pre">PIL.Image</span></code>, <code class="docutils literal notranslate"><span class="pre">matplotlib.pyplot</span></code> (์ด๋ฏธ์ง ๋ถ๋ฌ์ค๊ธฐ ๋ฐ ์ด๋ฏธ์ง ํ์)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torchvision.transforms</span></code> (PIL ์ด๋ฏธ์ง๋ฅผ ํ
์๋ก ๋ณํ)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">torchvision.models</span></code> (๋ฏธ๋ฆฌ ํ์ต๋ ๋ชจ๋ธ ๋ถ๋ฌ์ค๊ธฐ ๋ฐ ํ์ต)</p></li>
<li><p><code class="docutils literal notranslate"><span class="pre">copy</span></code> (๋ชจ๋ธ์ ๋ณต์ฌ; ์์คํ
ํจํค์ง)</p></li>
</ul>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="kn">from</span> <span class="nn">__future__</span> <span class="kn">import</span> <span class="n">print_function</span>
<span class="kn">import</span> <span class="nn">torch</span>
<span class="kn">import</span> <span class="nn">torch.nn</span> <span class="k">as</span> <span class="nn">nn</span>
<span class="kn">import</span> <span class="nn">torch.nn.functional</span> <span class="k">as</span> <span class="nn">F</span>
<span class="kn">import</span> <span class="nn">torch.optim</span> <span class="k">as</span> <span class="nn">optim</span>
<span class="kn">from</span> <span class="nn">PIL</span> <span class="kn">import</span> <span class="n">Image</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">import</span> <span class="nn">torchvision.transforms</span> <span class="k">as</span> <span class="nn">transforms</span>
<span class="kn">import</span> <span class="nn">torchvision.models</span> <span class="k">as</span> <span class="nn">models</span>
<span class="kn">import</span> <span class="nn">copy</span>
</pre></div>
</div>
<p>๋ค์์ผ๋ก, ๋คํธ์ํฌ๋ฅผ ์คํ ํ ์ฅ์น๋ฅผ ์ ํํ๊ณ Style ์ด๋ฏธ์ง์ Content ์ด๋ฏธ์ง๋ฅผ ๊ฐ์ ธ์์ผ ํฉ๋๋ค.
ํฐ ์ด๋ฏธ์ง๋ก ๋ด๋ด ๋ณํ(neural transfer) ์๊ณ ๋ฆฌ์ฆ์ ์คํํ๋ฉด ์๊ฐ์ด ์ค๋ ๊ฑธ๋ฆฌ๊ณ
GPU์์ ์คํํ ๋ ํจ์ฌ ๋นจ๋ผ์ง๋๋ค.
<code class="docutils literal notranslate"><span class="pre">torch.cuda.is_available()</span></code> ๋ฅผ ์ฌ์ฉํ์ฌ GPU๋ฅผ ์ด์ฉํ ์ ์๋์ง ์ ์ ์์ต๋๋ค.
๋ค์์ผ๋ก, ํํ ๋ฆฌ์ผ ์ ์ฒด์์ ์ฌ์ฉ ํ <code class="docutils literal notranslate"><span class="pre">torch.device</span></code> ๋ฅผ ์ค์ ํฉ๋๋ค.
๋ํ <code class="docutils literal notranslate"><span class="pre">.to(device)</span></code> ๋ฉ์๋๋ ํ
์ ๋๋ ๋ชจ๋์ ์ํ๋ ์ฅ์น๋ก ์ด๋ํ๋๋ฐ ์ฌ์ฉ๋ฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></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="s2">"cuda"</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="s2">"cpu"</span><span class="p">)</span>
</pre></div>
</div>
</div>
<div class="section" id="id5">
<h2>์ด๋ฏธ์ง ๋ถ๋ฌ์ค๊ธฐ<a class="headerlink" href="#id5" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>์ง๊ธ Style, Content ์ด๋ฏธ์ง๋ฅผ ๊ฐ์ ธ์ฌ ๊ฒ์
๋๋ค. ์๋ณธ PIL ์ด๋ฏธ์ง๋ 0๊ณผ 255 ์ฌ์ด์ ๊ฐ์ ๊ฐ์ง๋ง,
ํ
์๋ก ๋ณํ๋ ๋ 0์์ 1์ฌ์ด๋ก ๋ณํ๋ฉ๋๋ค.
์ด๋ฏธ์ง๋ ๋์ผํ ์ฐจ์์ ๊ฐ์ง๋๋ก ํฌ๊ธฐ๊ฐ ๋ณํ๋์ด์ผ ํฉ๋๋ค.
์ฃผ๋ชฉํด์ผ ํ ์ค์ํ ์ธ๋ถ ์ฌํญ์ torch ๋ผ์ด๋ธ๋ฌ๋ฆฌ์ ์ ๊ฒฝ๋ง์ 0์์ 1์ฌ์ด์ ํ
์ ๊ฐ์ผ๋ก ํ๋ จ๋๋ค๋ ๊ฒ์
๋๋ค.
๋ง์ฝ 0์์ 255๊ฐ์ ๊ฐ์ง๋ ํ
์ ์ด๋ฏธ์ง๊ฐ ๋คํธ์ํฌ์ ์
๋ ฅ๋๋ ๊ฒฝ์ฐ,
ํ์ฑํ๋ ํน์ง ๋งต์ด Content์ Style์ ๊ฐ์งํ ์ ์์ต๋๋ค.
๊ทธ๋ฌ๋, Caffe ๋ผ์ด๋ธ๋ฌ๋ฆฌ์ ๋ฏธ๋ฆฌ ํ์ต๋ ๋คํธ์ํฌ๋ 0์์ 255๊ฐ์ ๊ฐ์ง๋ ์
๋ ฅ์ผ๋ก ํ๋ จ๋ฉ๋๋ค.</p>
<div class="admonition note">
<p class="admonition-title">์ฐธ๊ณ </p>
<p>ํํ ๋ฆฌ์ผ์ ์คํํ๋๋ฐ ํ์ํ ์ด๋ฏธ์ง๋ฅผ ๋ค์ด๋ก๋ ํ ์ ์๋ ์ฃผ์๋ ๋ค์๊ณผ ๊ฐ์ต๋๋ค.
<a class="reference external" href="https://tutorials.pytorch.kr/_static/img/neural-style/picasso.jpg">picasso.jpg</a> ์
<a class="reference external" href="https://tutorials.pytorch.kr/_static/img/neural-style/dancing.jpg">dancing.jpg</a>.
๋ ์ด๋ฏธ์ง๋ฅผ ๋ค์ด๋ก๋ํ๊ณ ํ์ฌ ์์
ํด๋์ <code class="docutils literal notranslate"><span class="pre">images</span></code> ํด๋์ ์ถ๊ฐํ์ธ์.</p>
</div>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># ์ํ๋ ์ถ๋ ฅ ์ด๋ฏธ์ง์ ํฌ๊ธฐ</span>
<span class="n">imsize</span> <span class="o">=</span> <span class="mi">512</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="mi">128</span> <span class="c1"># ๋ง์ฝ GPU๊ฐ ์์ ๊ฒฝ์ฐ ์์ ํฌ๊ธฐ๋ฅผ ์ฌ์ฉ</span>
<span class="n">loader</span> <span class="o">=</span> <span class="n">transforms</span><span class="o">.</span><span class="n">Compose</span><span class="p">([</span>
<span class="n">transforms</span><span class="o">.</span><span class="n">Resize</span><span class="p">(</span><span class="n">imsize</span><span class="p">),</span> <span class="c1"># ๊ฐ์ ธ์จ ์ด๋ฏธ์ง ํฌ๊ธฐ ์กฐ์ </span>
<span class="n">transforms</span><span class="o">.</span><span class="n">ToTensor</span><span class="p">()])</span> <span class="c1"># ํ
์๋ก ๋ณํ</span>
<span class="k">def</span> <span class="nf">image_loader</span><span class="p">(</span><span class="n">image_name</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">open</span><span class="p">(</span><span class="n">image_name</span><span class="p">)</span>
<span class="c1"># ๋คํธ์ํฌ์ ์
๋ ฅ ์ฐจ์์ ๋ง์ถ๊ธฐ ์ํด ํ์ํ ๊ฐ์ง ๋ฐฐ์น ์ฐจ์</span>
<span class="n">image</span> <span class="o">=</span> <span class="n">loader</span><span class="p">(</span><span class="n">image</span><span class="p">)</span><span class="o">.</span><span class="n">unsqueeze</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span>
<span class="k">return</span> <span class="n">image</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">torch</span><span class="o">.</span><span class="n">float</span><span class="p">)</span>
<span class="n">style_img</span> <span class="o">=</span> <span class="n">image_loader</span><span class="p">(</span><span class="s2">"./data/images/neural-style/picasso.jpg"</span><span class="p">)</span>
<span class="n">content_img</span> <span class="o">=</span> <span class="n">image_loader</span><span class="p">(</span><span class="s2">"./data/images/neural-style/dancing.jpg"</span><span class="p">)</span>
<span class="k">assert</span> <span class="n">style_img</span><span class="o">.</span><span class="n">size</span><span class="p">()</span> <span class="o">==</span> <span class="n">content_img</span><span class="o">.</span><span class="n">size</span><span class="p">(),</span> \
<span class="s2">"we need to import style and content images of the same size"</span>
</pre></div>
</div>
<p>์ด์ ์ด๋ฏธ์ง๋ฅผ PIL ํ์์ผ๋ก ๋ค์ ๋ณํํ๊ณ <code class="docutils literal notranslate"><span class="pre">plt.imshow</span></code> ๋ฅผ
์ฌ์ฉํด ์ด๋ฏธ์ง๋ฅผ ํ์ํ๋ ํจ์๋ฅผ ๋ง๋ค๊ฒ ์ต๋๋ค.
Content์ Style ์ด๋ฏธ์ง๋ฅผ ํ์ํ์ฌ ์ฌ๋ฐ๋ฅด๊ฒ ๊ฐ์ ธ์๋์ง ํ์ธํฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">unloader</span> <span class="o">=</span> <span class="n">transforms</span><span class="o">.</span><span class="n">ToPILImage</span><span class="p">()</span> <span class="c1"># PIL ์ด๋ฏธ์ง๋ก ๋ค์ ๋ณํ</span>
<span class="n">plt</span><span class="o">.</span><span class="n">ion</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">imshow</span><span class="p">(</span><span class="n">tensor</span><span class="p">,</span> <span class="n">title</span><span class="o">=</span><span class="kc">None</span><span class="p">):</span>
<span class="n">image</span> <span class="o">=</span> <span class="n">tensor</span><span class="o">.</span><span class="n">cpu</span><span class="p">()</span><span class="o">.</span><span class="n">clone</span><span class="p">()</span> <span class="c1"># ํ
์๋ฅผ ๋ณต์ ํ์ฌ ๋ณ๊ฒฝํ์ง ์์</span>
<span class="n">image</span> <span class="o">=</span> <span class="n">image</span><span class="o">.</span><span class="n">squeeze</span><span class="p">(</span><span class="mi">0</span><span class="p">)</span> <span class="c1"># ๊ฐ์ง ๋ฐฐ์น ์ฐจ์ ์ ๊ฑฐ</span>
<span class="n">image</span> <span class="o">=</span> <span class="n">unloader</span><span class="p">(</span><span class="n">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="p">)</span>
<span class="k">if</span> <span class="n">title</span> <span class="ow">is</span> <span class="ow">not</span> <span class="kc">None</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="n">title</span><span class="p">)</span>
<span class="n">plt</span><span class="o">.</span><span class="n">pause</span><span class="p">(</span><span class="mf">0.001</span><span class="p">)</span> <span class="c1"># plots๊ฐ ์
๋ฐ์ดํธ ๋๋๋ก ์ ์ ๋ฉ์ถค</span>
<span class="n">plt</span><span class="o">.</span><span class="n">figure</span><span class="p">()</span>
<span class="n">imshow</span><span class="p">(</span><span class="n">style_img</span><span class="p">,</span> <span class="n">title</span><span class="o">=</span><span class="s1">'Style Image'</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">imshow</span><span class="p">(</span><span class="n">content_img</span><span class="p">,</span> <span class="n">title</span><span class="o">=</span><span class="s1">'Content Image'</span><span class="p">)</span>
</pre></div>
</div>
<ul class="sphx-glr-horizontal">
<li><img src="../_images/sphx_glr_neural_style_tutorial_001.png" srcset="../_images/sphx_glr_neural_style_tutorial_001.png" alt="Style Image" class = "sphx-glr-multi-img"/></li>
<li><img src="../_images/sphx_glr_neural_style_tutorial_002.png" srcset="../_images/sphx_glr_neural_style_tutorial_002.png" alt="Content Image" class = "sphx-glr-multi-img"/></li>
</ul>
</div>
<div class="section" id="id6">
<h2>์์ค ํจ์<a class="headerlink" href="#id6" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<div class="section" id="content-loss">
<h3>์ฝํ
์ธ ์์ค(Content Loss)<a class="headerlink" href="#content-loss" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>Content ์์ค์ ๊ฐ ๊ณ์ธต์ ๋ํ Content ๊ฑฐ๋ฆฌ์ ๊ฐ์ค์น ๋ฒ์ ์ ๋ํ๋
๋๋ค.
์ด ํจ์๋ ์
๋ ฅ <span class="math">\(X\)</span> ๋ฅผ ์ฒ๋ฆฌํ๋ ๋ ์ด์ด <span class="math">\(L\)</span> ์ ํน์ง ๋งต <span class="math">\(F_{XL}\)</span> ์ ๊ฐ์ ธ์์
์ด๋ฏธ์ง <span class="math">\(X\)</span> ์ Content ์ด๋ฏธ์ง <span class="math">\(C\)</span> ์ฌ์ด์
๊ฐ์ค ์ฝํ
์ธ ๊ฑฐ๋ฆฌ(weighted content distance) <span class="math">\(w_{CL}.D_C^L(X,C)\)</span> ๋ฅผ ๋ฐํํฉ๋๋ค.
Content ๊ฑฐ๋ฆฌ๋ฅผ ๊ณ์ฐํ๊ธฐ ์ํด Content ์ด๋ฏธ์ง(<span class="math">\(F_{CL}\)</span>)์ ํน์ง ๋งต์ ํจ์์์ ์๊ณ ์์ด์ผํฉ๋๋ค.
<span class="math">\(F_{CL}\)</span> ์ ์
๋ ฅ์ผ๋ก ์ฌ์ฉํ๋ ์์ฑ์๋ฅผ ์ฌ์ฉํด ์ด ํจ์๋ฅผ torch ๋ชจ๋๋ก ๊ตฌํํฉ๋๋ค.
๊ฑฐ๋ฆฌ <span class="math">\(\|F_{XL} - F_{CL}\|^2\)</span> ๋ ๋ ๊ฐ์ ํน์ง ๋งต ์งํฉ์ ํ๊ท ์ ๊ณฑ ์ค์ฐจ์ด๋ฉฐ <code class="docutils literal notranslate"><span class="pre">nn.MSELoss</span></code> ๋ฅผ ์ฌ์ฉํ์ฌ ๊ณ์ฐํ ์ ์์ต๋๋ค.</p>
<p>Content ๊ฑฐ๋ฆฌ๋ฅผ ๊ณ์ฐํ๊ธฐ ์ํด ์ฌ์ฉ๋๋ ํฉ์ฑ๊ณฑ ๊ณ์ธต ๋ฐ๋ก ๋ค์ Content ์์ค ๋ชจ๋์ ์ถ๊ฐํฉ๋๋ค.
์ด๋ ๊ฒ ํ๋ฉด ์
๋ ฅ ์ด๋ฏธ์ง๊ฐ ์
๋ ฅ๋ ๋๋ง๋ค Content ์์ค์ด ์ํ๋ ๋ ์ด์ด์์
๊ณ์ฐ๋๊ณ autograd ์ ํตํด ๋ชจ๋ ๊ธฐ์ธ๊ธฐ๊ฐ ๊ณ์ฐ๋ฉ๋๋ค.
์ด์ Content ์์ค ๊ณ์ธต์ ๋ง๋ค๊ธฐ ์ํด Content ์์ค์ ๊ณ์ฐํ ๋ค์ ๊ณ์ธต์ ์
๋ ฅ์ ๋ฐํํ๋
<code class="docutils literal notranslate"><span class="pre">forward</span></code> ๋ฉ์๋๋ฅผ ์ ์ํด์ผํฉ๋๋ค.
๊ณ์ฐ๋ ์์ค์ ๋ชจ๋์ ๋งค๊ฐ ๋ณ์๋ก์จ ์ ์ฅ๋ฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">ContentLoss</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</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">target</span><span class="p">,):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">ContentLoss</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="c1"># ๊ธฐ์ธ๊ธฐ๋ฅผ ๋์ ์ผ๋ก ๊ณ์ฐํ๋๋ฐ ์ฌ์ฉ๋๋ tree๋ก๋ถํฐ ํ๊น Content๋ฅผ `๋ถ๋ฆฌ(detach)` ํฉ๋๋ค.</span>
<span class="c1"># ์ด๊ฒ์ ๋ณ์๊ฐ ์๋๋ผ ๋ช
์๋ ๊ฐ์
๋๋ค.</span>
<span class="c1"># ๊ทธ๋ ์ง ์์ผ๋ฉด criterion์ forward ๋ฉ์๋์์ ์ค๋ฅ๊ฐ ๋ฐ์ํฉ๋๋ค.</span>
<span class="bp">self</span><span class="o">.</span><span class="n">target</span> <span class="o">=</span> <span class="n">target</span><span class="o">.</span><span class="n">detach</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">):</span>
<span class="bp">self</span><span class="o">.</span><span class="n">loss</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">mse_loss</span><span class="p">(</span><span class="nb">input</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">target</span><span class="p">)</span>
<span class="k">return</span> <span class="nb">input</span>
</pre></div>
</div>
<div class="admonition note">
<p class="admonition-title">์ฐธ๊ณ </p>
<p><strong>์ค์ ์ธ๋ถ ์ฌํญ</strong>: ๋ชจ๋์ ์ด๋ฆ์ <code class="docutils literal notranslate"><span class="pre">ContentLoss</span></code> ์ง๋ง, ์ค์ PyTorch์ ์์ค ํจ์๋ ์๋๋๋ค.
๋ง์ฝ Content ์์ค์ PyTorch์ ์์ค ํจ์๋ก ์ ์ํ๋ ค๋ฉด Pytorch์ autograd ํจ์๋ฅผ ์์ฑํ์ฌ
<code class="docutils literal notranslate"><span class="pre">backward</span></code> ๋ฉ์๋์์ ์๋์ผ๋ก ๊ธฐ์ธ๊ธฐ๋ฅผ ๋ค์ ๊ณ์ฐ ๋ฐ ๊ตฌํ ํด์ผ ํฉ๋๋ค.</p>
</div>
</div>
<div class="section" id="style-loss">
<h3>์คํ์ผ ์์ค(Style Loss)<a class="headerlink" href="#style-loss" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h3>
<p>Style ์์ค ๋ชจ๋์ Content ์์ค ๋ชจ๋๊ณผ ์ ์ฌํ๊ฒ ๊ตฌํ๋ฉ๋๋ค.
๋คํธ์ํฌ์์ ํด๋น ๊ณ์ธต์ Style ์์ค์ ๊ณ์ฐํ๋ ์ญํ ์ ํฉ๋๋ค.
Style ์์ค์ ๊ณ์ฐํ๊ธฐ ์ํด์, gram ํ๋ ฌ <span class="math">\(G_{XL}\)</span> ๋ฅผ ๊ณ์ฐํด์ผํฉ๋๋ค.
gram ํ๋ ฌ์ ์ฃผ์ด์ง ํ๋ ฌ์ ์ ์น ํ๋ ฌ์ ๊ณฑํ ๊ฒฐ๊ณผ์
๋๋ค.
์ด ์ดํ๋ฆฌ์ผ์ด์
์์ ์ฃผ์ด์ง ํ๋ ฌ์ ๊ณ์ธต <span class="math">\(L\)</span> ์ ํน์ง ๋งต <span class="math">\(F_{XL}\)</span> ์ ์ฌ๊ตฌ์ฑ๋ ๋ฒ์ ์
๋๋ค.
<span class="math">\(F_{XL}\)</span> ๋ <span class="math">\(K\)</span>x<span class="math">\(N\)</span> ํ๋ ฌ์ธ <span class="math">\(\hat{F}_{XL}\)</span> ์ ํ์ฑํ๋๋ก ์ฌ๊ตฌ์ฑ๋๋ฉฐ,
<span class="math">\(K\)</span> ๋ ๊ณ์ธต <span class="math">\(L\)</span> ์ ํน์ง ๋งต์ ์์ด๊ณ <span class="math">\(N\)</span>hat{F}_{XL}` ์ ์ฒซ ๋ฒ์งธ ์ค์ ๋ฒกํฐํ๋ ํน์ง ๋งต <span class="math">\(F_{XL}^1\)</span> ์ ํด๋นํฉ๋๋ค.</p>
<p>๋ง์ง๋ง์ผ๋ก, gram ํ๋ ฌ์ ํ๋ ฌ์์ ๊ฐ ๊ฐ์ ๊ฐ์ ์ด ๊ฐ์๋ก ๋๋์ด ์ ๊ทํ ํด์ผ ํฉ๋๋ค.
์ ๊ทํ๋ ํฐ <span class="math">\(N\)</span> ์ฐจ์์ ๊ฐ์ง๋ <span class="math">\(\hat{F}_{XL}\)</span> ํ๋ ฌ์ด gram ํ๋ ฌ์์ ๋ ํฐ ๊ฐ์ ์์ฑํ๋ค๋ ์ฌ์ค์ ๋์ํ๊ธฐ ์ํ ๊ฒ์
๋๋ค.</p>
<p>์ด์ ๊ฐ์ ๋ ํฐ ๊ฐ์ ์ฒซ ๋ฒ์งธ ๊ณ์ธต(ํ๋ง ๊ณ์ธต ์ )์ด ๊ฒฝ์ฌ ํ๊ฐ ์ค์ ๋ ํฐ ์ํฅ์ ๋ฏธ์น๊ฒํฉ๋๋ค.</p>
<p>Style ํน์ง์ ๋คํธ์ํฌ์ ๊น์ ๊ณ์ธต์ ์๋ ๊ฒฝํฅ์ด ์๊ธฐ์ ์ ๊ทํ ๋จ๊ณ๊ฐ ์ค์ํฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">gram_matrix</span><span class="p">(</span><span class="nb">input</span><span class="p">):</span>
<span class="n">a</span><span class="p">,</span> <span class="n">b</span><span class="p">,</span> <span class="n">c</span><span class="p">,</span> <span class="n">d</span> <span class="o">=</span> <span class="nb">input</span><span class="o">.</span><span class="n">size</span><span class="p">()</span> <span class="c1"># a=๋ฐฐ์น ํฌ๊ธฐ(=1)</span>
<span class="c1"># b=ํน์ง ๋งต์ ์</span>
<span class="c1"># (c,d)=ํน์ง ๋งต์ ์ฐจ์ (N=c*d)</span>
<span class="n">features</span> <span class="o">=</span> <span class="nb">input</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="n">a</span> <span class="o">*</span> <span class="n">b</span><span class="p">,</span> <span class="n">c</span> <span class="o">*</span> <span class="n">d</span><span class="p">)</span> <span class="c1"># F_XL์ \hat F_XL๋ก ํฌ๊ธฐ ์กฐ์ </span>
<span class="n">G</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">mm</span><span class="p">(</span><span class="n">features</span><span class="p">,</span> <span class="n">features</span><span class="o">.</span><span class="n">t</span><span class="p">())</span> <span class="c1"># gram product๋ฅผ ๊ณ์ฐ</span>
<span class="c1"># ๊ฐ ํน์ง ๋งต์ด ๊ฐ๋ ๊ฐ์ ์๋ก ๋๋์ด</span>
<span class="c1"># gram ํ๋ ฌ์ ๊ฐ์ '์ ๊ทํ'</span>
<span class="k">return</span> <span class="n">G</span><span class="o">.</span><span class="n">div</span><span class="p">(</span><span class="n">a</span> <span class="o">*</span> <span class="n">b</span> <span class="o">*</span> <span class="n">c</span> <span class="o">*</span> <span class="n">d</span><span class="p">)</span>
</pre></div>
</div>
<p>์ด์ Style ์์ค ๋ชจ๋์ Content ์์ค ๋ชจ๋๊ณผ ๊ฑฐ์ ๋๊ฐ์ ๋ณด์
๋๋ค.
Style ๊ฑฐ๋ฆฌ๋ <span class="math">\(G_{XL}\)</span> ์ <span class="math">\(G_{SL}\)</span> ์ฌ์ด์ ํ๊ท ์ ๊ณฑ ์ค์ฐจ๋ฅผ ์ฌ์ฉํด ๊ณ์ฐ๋ฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">class</span> <span class="nc">StyleLoss</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</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">target_feature</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">StyleLoss</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="bp">self</span><span class="o">.</span><span class="n">target</span> <span class="o">=</span> <span class="n">gram_matrix</span><span class="p">(</span><span class="n">target_feature</span><span class="p">)</span><span class="o">.</span><span class="n">detach</span><span class="p">()</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="nb">input</span><span class="p">):</span>
<span class="n">G</span> <span class="o">=</span> <span class="n">gram_matrix</span><span class="p">(</span><span class="nb">input</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">loss</span> <span class="o">=</span> <span class="n">F</span><span class="o">.</span><span class="n">mse_loss</span><span class="p">(</span><span class="n">G</span><span class="p">,</span> <span class="bp">self</span><span class="o">.</span><span class="n">target</span><span class="p">)</span>
<span class="k">return</span> <span class="nb">input</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="id7">
<h2>๋ชจ๋ธ ๊ฐ์ ธ์ค๊ธฐ<a class="headerlink" href="#id7" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>์ด์ ๋ ์ฌ์ ํ๋ จ๋ ์ ๊ฒฝ๋ง์ ๊ฐ์ ธ์์ผ ํฉ๋๋ค.
๋
ผ๋ฌธ์์ ์ฌ์ฉ๋ ๊ฒ๊ณผ ๊ฐ์ด 19 ๊ณ์ธต์ ๊ฐ์ง VGG ๋คํธ์ํฌ๋ฅผ ์ฌ์ฉํ ๊ฒ์
๋๋ค.</p>
<p>PyTorch์ VGG ๊ตฌํ์ <code class="docutils literal notranslate"><span class="pre">features</span></code> (ํฉ์ฑ๊ณฑ ๋ฐ ํ๋ง ๊ณ์ธต ํฌํจ) ์
<code class="docutils literal notranslate"><span class="pre">classifier</span></code> (์์ ์ฐ๊ฒฐ ๊ณ์ธต ํฌํจ) ๋ ๊ฐ์ง ํ์ <code class="docutils literal notranslate"><span class="pre">Sequential</span></code> ๋ชจ๋๋ก ๊ตฌ๋ถ ๋ ๋ชจ๋์
๋๋ค.
Content ๋ฐ Style ์์ค์ ์ธก์ ํ๊ธฐ ์ํด์ ๊ฐ ํฉ์ฑ๊ณฑ ๊ณ์ธต์ ์ถ๋ ฅ์ด ํ์ํ๊ธฐ ๋๋ฌธ์ <code class="docutils literal notranslate"><span class="pre">features</span></code> ๋ชจ๋์ ์ฌ์ฉํ ๊ฒ์
๋๋ค.
์ผ๋ถ ๊ณ์ธต์ ํ๋ จํ๋ ์ค ํ๊ฐ์ ๋ค๋ฅธ ๋์์ ํ๋ฏ๋ก, ๋คํธ์ํฌ๋ฅผ <code class="docutils literal notranslate"><span class="pre">.eval()</span></code> ๋ฅผ ์ฌ์ฉํด ํ๊ฐ ๋ชจ๋๋ก ์ค์ ํด์ผํฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">cnn</span> <span class="o">=</span> <span class="n">models</span><span class="o">.</span><span class="n">vgg19</span><span class="p">(</span><span class="n">pretrained</span><span class="o">=</span><span class="kc">True</span><span class="p">)</span><span class="o">.</span><span class="n">features</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="o">.</span><span class="n">eval</span><span class="p">()</span>
</pre></div>
</div>
<div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Downloading: "https://download.pytorch.org/models/vgg19-dcbb9e9d.pth" to /root/.cache/torch/hub/checkpoints/vgg19-dcbb9e9d.pth
0%| | 0.00/548M [00:00<?, ?B/s]
1%|1 | 5.88M/548M [00:00<00:10, 56.0MB/s]
2%|2 | 12.3M/548M [00:00<00:08, 62.5MB/s]
3%|3 | 18.5M/548M [00:00<00:08, 63.3MB/s]
4%|4 | 24.5M/548M [00:00<00:08, 61.8MB/s]
6%|5 | 30.8M/548M [00:00<00:08, 63.0MB/s]
7%|6 | 36.8M/548M [00:00<00:15, 35.7MB/s]
8%|7 | 43.1M/548M [00:00<00:12, 42.1MB/s]
9%|9 | 50.2M/548M [00:01<00:10, 49.7MB/s]
10%|# | 57.4M/548M [00:01<00:09, 56.2MB/s]
12%|#1 | 65.2M/548M [00:01<00:08, 62.7MB/s]
13%|#3 | 72.4M/548M [00:01<00:07, 66.3MB/s]
14%|#4 | 79.3M/548M [00:01<00:07, 67.2MB/s]
16%|#5 | 86.1M/548M [00:01<00:07, 63.9MB/s]
17%|#7 | 93.8M/548M [00:01<00:06, 68.5MB/s]
19%|#8 | 101M/548M [00:01<00:06, 72.0MB/s]
20%|#9 | 109M/548M [00:01<00:06, 73.2MB/s]
21%|##1 | 116M/548M [00:01<00:06, 70.6MB/s]
22%|##2 | 123M/548M [00:02<00:06, 69.8MB/s]
24%|##3 | 129M/548M [00:02<00:06, 68.7MB/s]
25%|##4 | 136M/548M [00:02<00:06, 68.4MB/s]
26%|##6 | 143M/548M [00:02<00:06, 65.2MB/s]
27%|##7 | 149M/548M [00:02<00:07, 59.4MB/s]
28%|##8 | 155M/548M [00:02<00:06, 61.4MB/s]
30%|##9 | 162M/548M [00:02<00:06, 64.7MB/s]
31%|### | 169M/548M [00:02<00:06, 65.9MB/s]
32%|###1 | 175M/548M [00:02<00:05, 66.0MB/s]
33%|###3 | 182M/548M [00:03<00:05, 66.2MB/s]
35%|###4 | 189M/548M [00:03<00:05, 70.3MB/s]
36%|###5 | 196M/548M [00:03<00:05, 69.7MB/s]
37%|###6 | 203M/548M [00:03<00:05, 69.5MB/s]
38%|###8 | 210M/548M [00:03<00:05, 70.7MB/s]
40%|###9 | 218M/548M [00:03<00:04, 74.5MB/s]
41%|####1 | 225M/548M [00:03<00:04, 70.1MB/s]
42%|####2 | 233M/548M [00:03<00:04, 73.3MB/s]
44%|####3 | 240M/548M [00:03<00:04, 75.2MB/s]
45%|####5 | 247M/548M [00:03<00:04, 75.0MB/s]
46%|####6 | 255M/548M [00:04<00:04, 73.9MB/s]
48%|####7 | 262M/548M [00:04<00:04, 73.6MB/s]
49%|####9 | 269M/548M [00:04<00:03, 75.2MB/s]
50%|##### | 276M/548M [00:04<00:03, 72.2MB/s]
52%|#####1 | 283M/548M [00:04<00:03, 71.8MB/s]
53%|#####2 | 290M/548M [00:04<00:03, 70.0MB/s]
54%|#####4 | 297M/548M [00:04<00:03, 69.8MB/s]
56%|#####5 | 305M/548M [00:04<00:03, 73.3MB/s]
57%|#####6 | 312M/548M [00:04<00:03, 71.1MB/s]
58%|#####8 | 319M/548M [00:05<00:03, 72.3MB/s]
59%|#####9 | 326M/548M [00:05<00:03, 69.8MB/s]
61%|###### | 333M/548M [00:05<00:03, 69.2MB/s]
62%|######2 | 340M/548M [00:05<00:03, 71.4MB/s]
63%|######3 | 347M/548M [00:05<00:02, 72.7MB/s]
65%|######4 | 354M/548M [00:05<00:02, 71.4MB/s]
66%|######5 | 361M/548M [00:05<00:02, 72.8MB/s]
67%|######7 | 368M/548M [00:05<00:02, 72.3MB/s]
68%|######8 | 375M/548M [00:05<00:02, 72.5MB/s]
70%|######9 | 382M/548M [00:05<00:02, 72.8MB/s]
71%|#######1 | 390M/548M [00:06<00:02, 75.3MB/s]
73%|#######2 | 397M/548M [00:06<00:02, 75.5MB/s]
74%|#######3 | 405M/548M [00:06<00:01, 76.0MB/s]
75%|#######5 | 412M/548M [00:06<00:01, 76.7MB/s]
77%|#######6 | 420M/548M [00:06<00:01, 71.9MB/s]
78%|#######7 | 426M/548M [00:06<00:01, 68.7MB/s]
79%|#######9 | 433M/548M [00:06<00:01, 69.3MB/s]
80%|######## | 440M/548M [00:06<00:01, 70.4MB/s]
82%|########1 | 448M/548M [00:06<00:01, 72.2MB/s]
83%|########2 | 454M/548M [00:07<00:01, 72.4MB/s]
84%|########4 | 462M/548M [00:07<00:01, 74.0MB/s]
86%|########5 | 469M/548M [00:07<00:01, 72.8MB/s]
87%|########6 | 476M/548M [00:07<00:01, 68.1MB/s]
88%|########8 | 483M/548M [00:07<00:01, 66.2MB/s]
89%|########9 | 489M/548M [00:07<00:00, 66.7MB/s]
90%|######### | 496M/548M [00:07<00:00, 67.5MB/s]
92%|#########1| 504M/548M [00:07<00:00, 72.0MB/s]
93%|#########3| 511M/548M [00:07<00:00, 72.1MB/s]
94%|#########4| 517M/548M [00:07<00:00, 71.5MB/s]
96%|#########5| 524M/548M [00:08<00:00, 69.9MB/s]
97%|#########7| 532M/548M [00:08<00:00, 72.3MB/s]
98%|#########8| 539M/548M [00:08<00:00, 72.4MB/s]
100%|#########9| 546M/548M [00:08<00:00, 74.9MB/s]
100%|##########| 548M/548M [00:08<00:00, 68.6MB/s]
</pre></div>
</div>
<p>์ถ๊ฐ์ ์ผ๋ก, VGG ๋คํธ์ํฌ๋
mean=[0.485, 0.456, 0.406] ์ std=[0.229, 0.224, 0.225]๋ก ๊ฐ ์ฑ๋์ด ์ ๊ทํ ๋
์ด๋ฏธ์ง๋ก ํ๋ จ๋ฉ๋๋ค.
์ด๋ฏธ์ง๋ฅผ ๋คํธ์ํฌ๋ก ์
๋ ฅํ๊ธฐ ์ ์ ์ ๊ทํํ๋๋ฐ ์ฌ์ฉํฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">cnn_normalization_mean</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">([</span><span class="mf">0.485</span><span class="p">,</span> <span class="mf">0.456</span><span class="p">,</span> <span class="mf">0.406</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">cnn_normalization_std</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">([</span><span class="mf">0.229</span><span class="p">,</span> <span class="mf">0.224</span><span class="p">,</span> <span class="mf">0.225</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="c1"># ์
๋ ฅ ์ด๋ฏธ์ง๋ฅผ ์ ๊ทํํ๋ ๋ชจ๋์ ์์ฑํ์ฌ ์ฝ๊ฒ ``nn.Sequential`` ์ ๋ฃ์ ์ ์์ต๋๋ค.</span>
<span class="k">class</span> <span class="nc">Normalization</span><span class="p">(</span><span class="n">nn</span><span class="o">.</span><span class="n">Module</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">mean</span><span class="p">,</span> <span class="n">std</span><span class="p">):</span>
<span class="nb">super</span><span class="p">(</span><span class="n">Normalization</span><span class="p">,</span> <span class="bp">self</span><span class="p">)</span><span class="o">.</span><span class="fm">__init__</span><span class="p">()</span>
<span class="c1"># .view ๋ mean๊ณผ std์ ํ์ธํด [B x C x H x W] ํํ์</span>
<span class="c1"># ์ด๋ฏธ์ง ํ
์๋ฅผ ์ง์ ์ ์ผ๋ก ์์
ํ ์ ์๋๋ก [C x 1 x 1] ํํ๋ก ๋ง๋ญ๋๋ค.</span>
<span class="c1"># B๋ ๋ฐฐ์น ํฌ๊ธฐ์
๋๋ค. C๋ ์ฑ๋์ ์์
๋๋ค. H๋ ๋์ด๊ณ W๋ ๋๋น์
๋๋ค.</span>
<span class="bp">self</span><span class="o">.</span><span class="n">mean</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="n">mean</span><span class="p">)</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
<span class="bp">self</span><span class="o">.</span><span class="n">std</span> <span class="o">=</span> <span class="n">torch</span><span class="o">.</span><span class="n">tensor</span><span class="p">(</span><span class="n">std</span><span class="p">)</span><span class="o">.</span><span class="n">view</span><span class="p">(</span><span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
<span class="k">def</span> <span class="nf">forward</span><span class="p">(</span><span class="bp">self</span><span class="p">,</span> <span class="n">img</span><span class="p">):</span>
<span class="c1"># ``img`` ์ ๊ทํ</span>
<span class="k">return</span> <span class="p">(</span><span class="n">img</span> <span class="o">-</span> <span class="bp">self</span><span class="o">.</span><span class="n">mean</span><span class="p">)</span> <span class="o">/</span> <span class="bp">self</span><span class="o">.</span><span class="n">std</span>
</pre></div>
</div>
<p><code class="docutils literal notranslate"><span class="pre">Sequential</span></code> ๋ชจ๋์ ์์๊ฐ ์๋ ํ์ ๋ชจ๋์ ๋ฆฌ์คํธ๊ฐ ํฌํจ๋ฉ๋๋ค.
์๋ฅผ ๋ค์ด, <code class="docutils literal notranslate"><span class="pre">vgg19.features</span></code> ์ ์ฌ๋ฐ๋ฅธ ์์๋ก ์ ๋ ฌ ๋
์ํ์ค( <code class="docutils literal notranslate"><span class="pre">Conv2d</span></code>, <code class="docutils literal notranslate"><span class="pre">ReLU</span></code>, <code class="docutils literal notranslate"><span class="pre">MaxPool2d</span></code>, <code class="docutils literal notranslate"><span class="pre">Conv2d</span></code>, <code class="docutils literal notranslate"><span class="pre">ReLU</span></code>โฆ)๊ฐ ํฌํจ๋์ด ์์ต๋๋ค.
Content ์์ค๊ณผ Style ์์ค ๊ณ์ธต์ ๊ฐ์งํ๋ ํฉ์ฑ๊ณฑ ๊ณ์ธต ๋ฐ๋ก ๋ค์ ์ถ๊ฐํด์ผํฉ๋๋ค.
์ด๋ ๊ฒ ํ๊ธฐ ์ํด์๋ Content ์์ค๊ณผ Style ์์ค ๋ชจ๋์ด
์ฌ๋ฐ๋ฅด๊ฒ ์ฝ์
๋ ์๋ก์ด <code class="docutils literal notranslate"><span class="pre">Sequential</span></code> ๋ชจ๋์ ๋ง๋ค์ด์ผ ํฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="c1"># Style / Content ์์ค ๊ณ์ฐ์ ์ํ๋ ๊ณ์ธต</span>
<span class="n">content_layers_default</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'conv_4'</span><span class="p">]</span>
<span class="n">style_layers_default</span> <span class="o">=</span> <span class="p">[</span><span class="s1">'conv_1'</span><span class="p">,</span> <span class="s1">'conv_2'</span><span class="p">,</span> <span class="s1">'conv_3'</span><span class="p">,</span> <span class="s1">'conv_4'</span><span class="p">,</span> <span class="s1">'conv_5'</span><span class="p">]</span>
<span class="k">def</span> <span class="nf">get_style_model_and_losses</span><span class="p">(</span><span class="n">cnn</span><span class="p">,</span> <span class="n">normalization_mean</span><span class="p">,</span> <span class="n">normalization_std</span><span class="p">,</span>
<span class="n">style_img</span><span class="p">,</span> <span class="n">content_img</span><span class="p">,</span>
<span class="n">content_layers</span><span class="o">=</span><span class="n">content_layers_default</span><span class="p">,</span>
<span class="n">style_layers</span><span class="o">=</span><span class="n">style_layers_default</span><span class="p">):</span>
<span class="c1"># ๋ชจ๋ ์ ๊ทํ</span>
<span class="n">normalization</span> <span class="o">=</span> <span class="n">Normalization</span><span class="p">(</span><span class="n">normalization_mean</span><span class="p">,</span> <span class="n">normalization_std</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="c1"># Content / Style ์์ค์ด ๋ฐ๋ณต์ ์ผ๋ก ์ ๊ทผํ ์ ์๋๋ก ํ๊ธฐ ์ํด</span>
<span class="n">content_losses</span> <span class="o">=</span> <span class="p">[]</span>
<span class="n">style_losses</span> <span class="o">=</span> <span class="p">[]</span>
<span class="c1"># ``cnn`` ์ด ``nn.Sequential`` ์ด๋ผ๊ณ ๊ฐ์ ํ๊ณ ,</span>
<span class="c1"># ์์ฐจ์ ์ผ๋ก ํ์ฑํ๋์ด์ผ ํ๋ ๋ชจ๋์ ์๋ก์ด ``nn.Sequential`` ์ ๋ง๋ญ๋๋ค.</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">Sequential</span><span class="p">(</span><span class="n">normalization</span><span class="p">)</span>
<span class="n">i</span> <span class="o">=</span> <span class="mi">0</span> <span class="c1"># conv๋ฅผ ๋ณผ ๋๋ง๋ค ์ฆ๊ฐ</span>
<span class="k">for</span> <span class="n">layer</span> <span class="ow">in</span> <span class="n">cnn</span><span class="o">.</span><span class="n">children</span><span class="p">():</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">Conv2d</span><span class="p">):</span>
<span class="n">i</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="n">name</span> <span class="o">=</span> <span class="s1">'conv_</span><span class="si">{}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">i</span><span class="p">)</span>
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">):</span>
<span class="n">name</span> <span class="o">=</span> <span class="s1">'relu_</span><span class="si">{}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">i</span><span class="p">)</span>
<span class="c1"># ์๋์ ์ถ๊ฐํ ``ContentLoss`` ์ ``StyleLoss`` ๋ in-place ๋ฒ์ ์์๋ ์ ๋์ํ์ง ์์ต๋๋ค.</span>
<span class="c1"># ๊ทธ๋์ ์ฌ๊ธฐ์๋ out-of-place๋ก ๋์ฒดํฉ๋๋ค.</span>
<span class="n">layer</span> <span class="o">=</span> <span class="n">nn</span><span class="o">.</span><span class="n">ReLU</span><span class="p">(</span><span class="n">inplace</span><span class="o">=</span><span class="kc">False</span><span class="p">)</span>
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">MaxPool2d</span><span class="p">):</span>
<span class="n">name</span> <span class="o">=</span> <span class="s1">'pool_</span><span class="si">{}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">i</span><span class="p">)</span>
<span class="k">elif</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">layer</span><span class="p">,</span> <span class="n">nn</span><span class="o">.</span><span class="n">BatchNorm2d</span><span class="p">):</span>
<span class="n">name</span> <span class="o">=</span> <span class="s1">'bn_</span><span class="si">{}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">i</span><span class="p">)</span>
<span class="k">else</span><span class="p">:</span>
<span class="k">raise</span> <span class="ne">RuntimeError</span><span class="p">(</span><span class="s1">'Unrecognized layer: </span><span class="si">{}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">layer</span><span class="o">.</span><span class="vm">__class__</span><span class="o">.</span><span class="vm">__name__</span><span class="p">))</span>
<span class="n">model</span><span class="o">.</span><span class="n">add_module</span><span class="p">(</span><span class="n">name</span><span class="p">,</span> <span class="n">layer</span><span class="p">)</span>
<span class="k">if</span> <span class="n">name</span> <span class="ow">in</span> <span class="n">content_layers</span><span class="p">:</span>
<span class="c1"># Content ์์ค ์ถ๊ฐ</span>
<span class="n">target</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="n">content_img</span><span class="p">)</span><span class="o">.</span><span class="n">detach</span><span class="p">()</span>
<span class="n">content_loss</span> <span class="o">=</span> <span class="n">ContentLoss</span><span class="p">(</span><span class="n">target</span><span class="p">)</span>
<span class="n">model</span><span class="o">.</span><span class="n">add_module</span><span class="p">(</span><span class="s2">"content_loss_</span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">i</span><span class="p">),</span> <span class="n">content_loss</span><span class="p">)</span>
<span class="n">content_losses</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">content_loss</span><span class="p">)</span>
<span class="k">if</span> <span class="n">name</span> <span class="ow">in</span> <span class="n">style_layers</span><span class="p">:</span>
<span class="c1"># Style ์์ค ์ถ๊ฐ</span>
<span class="n">target_feature</span> <span class="o">=</span> <span class="n">model</span><span class="p">(</span><span class="n">style_img</span><span class="p">)</span><span class="o">.</span><span class="n">detach</span><span class="p">()</span>
<span class="n">style_loss</span> <span class="o">=</span> <span class="n">StyleLoss</span><span class="p">(</span><span class="n">target_feature</span><span class="p">)</span>
<span class="n">model</span><span class="o">.</span><span class="n">add_module</span><span class="p">(</span><span class="s2">"style_loss_</span><span class="si">{}</span><span class="s2">"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">i</span><span class="p">),</span> <span class="n">style_loss</span><span class="p">)</span>
<span class="n">style_losses</span><span class="o">.</span><span class="n">append</span><span class="p">(</span><span class="n">style_loss</span><span class="p">)</span>
<span class="c1"># ์ด์ ๋ง์ง๋ง Content ๋ฐ Style ์์ค ๋ค์ ๊ณ์ธต์ ์๋ผ๋
๋๋ค.</span>
<span class="k">for</span> <span class="n">i</span> <span class="ow">in</span> <span class="nb">range</span><span class="p">(</span><span class="nb">len</span><span class="p">(</span><span class="n">model</span><span class="p">)</span> <span class="o">-</span> <span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">,</span> <span class="o">-</span><span class="mi">1</span><span class="p">):</span>
<span class="k">if</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">[</span><span class="n">i</span><span class="p">],</span> <span class="n">ContentLoss</span><span class="p">)</span> <span class="ow">or</span> <span class="nb">isinstance</span><span class="p">(</span><span class="n">model</span><span class="p">[</span><span class="n">i</span><span class="p">],</span> <span class="n">StyleLoss</span><span class="p">):</span>
<span class="k">break</span>
<span class="n">model</span> <span class="o">=</span> <span class="n">model</span><span class="p">[:(</span><span class="n">i</span> <span class="o">+</span> <span class="mi">1</span><span class="p">)]</span>
<span class="k">return</span> <span class="n">model</span><span class="p">,</span> <span class="n">style_losses</span><span class="p">,</span> <span class="n">content_losses</span>
</pre></div>
</div>
<p>๋ค์์ผ๋ก ์
๋ ฅ ์ด๋ฏธ์ง๋ฅผ ์ ํํฉ๋๋ค. Content ์ด๋ฏธ์ง ์ฌ๋ณธ์ด๋ ๋ฐฑ์ ์ก์์ ์ฌ์ฉํ ์ ์์ต๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">input_img</span> <span class="o">=</span> <span class="n">content_img</span><span class="o">.</span><span class="n">clone</span><span class="p">()</span>
<span class="c1"># ๋ง์ฝ ํ์ดํธ ๋
ธ์ด์ฆ(white noise)์ ์ฌ์ฉํ๋ ค๋ฉด ์๋ ์ฃผ์์ ์ ๊ฑฐํ์ธ์:</span>
<span class="c1">#</span>
<span class="c1"># ::</span>
<span class="c1">#</span>
<span class="c1"># input_img = torch.randn(content_img.data.size(), device=device)</span>
<span class="c1"># ๊ทธ๋ฆผ์ ์๋ณธ ์
๋ ฅ ์ด๋ฏธ์ง๋ฅผ ์ถ๊ฐํฉ๋๋ค.</span>
<span class="n">plt</span><span class="o">.</span><span class="n">figure</span><span class="p">()</span>
<span class="n">imshow</span><span class="p">(</span><span class="n">input_img</span><span class="p">,</span> <span class="n">title</span><span class="o">=</span><span class="s1">'Input Image'</span><span class="p">)</span>
</pre></div>
</div>
<img src="../_images/sphx_glr_neural_style_tutorial_003.png" srcset="../_images/sphx_glr_neural_style_tutorial_003.png" alt="Input Image" class = "sphx-glr-single-img"/></div>
<div class="section" id="id8">
<h2>๊ฒฝ์ฌ ํ๊ฐ๋ฒ<a class="headerlink" href="#id8" title="์ด ์ ๋ชฉ์ ๋ํ ํผ๋จธ๋งํฌ">ยถ</a></h2>
<p>์๊ณ ๋ฆฌ์ฆ ์์ฑ์์ธ Leon Gatys๊ฐ
<a class="reference external" href="https://discuss.pytorch.org/t/pytorch-tutorial-for-neural-transfert-of-artistic-style/336/20?u=alexis-jacq">์ฌ๊ธฐ</a> ์์ ์ ์ํ๋ ๊ฒ์ฒ๋ผ,
L-BFGS ์๊ณ ๋ฆฌ์ฆ์ ์ฌ์ฉํ์ฌ ๊ฒฝ์ฌ ํ๊ฐ๋ฒ์ ์ํํฉ๋๋ค.
๋คํธ์ํฌ ํ๋ จ๊ณผ ๋ค๋ฅด๊ฒ Content / Style ์์ค์ ์ต์ํํ๊ธฐ ์ํด ์
๋ ฅ ์ด๋ฏธ์ง๋ฅผ ํ๋ จํ๋ ค๊ณ ํฉ๋๋ค.
ํ์ดํ ์น L-BFGS optimizer <code class="docutils literal notranslate"><span class="pre">optim.LBFGS</span></code> ๋ฅผ ๋ง๋ค๊ณ ์ต์ ํ ํ ํ
์๋ก ์ด๋ฏธ์ง๋ฅผ ์ ๋ฌํฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">get_input_optimizer</span><span class="p">(</span><span class="n">input_img</span><span class="p">):</span>
<span class="c1"># ์
๋ ฅ์ด ๊ธฐ์ธ๊ธฐ๊ฐ ํ์ํ ๋งค๊ฐ ๋ณ์์์ ํ์ํ๋ ์ค</span>
<span class="n">optimizer</span> <span class="o">=</span> <span class="n">optim</span><span class="o">.</span><span class="n">LBFGS</span><span class="p">([</span><span class="n">input_img</span><span class="p">])</span>
<span class="k">return</span> <span class="n">optimizer</span>
</pre></div>
</div>
<p>๋ง์ง๋ง์ผ๋ก ๋ด๋ด ๋ณํ(neural transfer)์ ์ํํ๋ ํจ์๋ฅผ ์ ์ํด์ผ ํฉ๋๋ค.
๋คํธ์ํฌ์ ๊ฐ ๋ฐ๋ณต ๋์, ์
๋ฐ์ดํธ๋ ์
๋ ฅ์ด ์ฃผ์ด์ง๊ณ ์๋ก์ด ์์ค์ ๊ณ์ฐํฉ๋๋ค.
๊ฐ ์์ค ๋ชจ๋(Loss module)์ <code class="docutils literal notranslate"><span class="pre">backward</span></code> ๋ฉ์๋๋ฅผ ์คํํ์ฌ ๊ธฐ์ธ๊ธฐ๋ฅผ ๋์ ์ผ๋ก ๊ณ์ฐํฉ๋๋ค.
optimizer๋ ๋ชจ๋์ ์ฌํ๊ฐํ๊ณ ์์ค์ ๋ฐํํ๋ โclosureโ ํจ์๊ฐ ํ์ํฉ๋๋ค.</p>
<p>์ฌ์ ํ ํด๊ฒฐํด์ผ ํ ๋ง์ง๋ง ์ ์ฝ์ด ์์ต๋๋ค.
๋คํธ์ํฌ๊ฐ ์ด๋ฏธ์ง์ ๋ํด 0 ~ 1 ํ
์ ๋ฒ์๋ฅผ ์ด๊ณผํ๋ ๊ฐ์ผ๋ก ์ต์ ํํ๋ ค ํ ์ ์์ต๋๋ค.
๋คํธ์ํฌ๊ฐ ์คํ๋ ๋๋ง๋ค ์
๋ ฅ๊ฐ์ 0์์ 1์ฌ์ด๋ก ์์ ํ์ฌ ๋ฌธ์ ๋ฅผ ํด๊ฒฐํ ์ ์์ต๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="k">def</span> <span class="nf">run_style_transfer</span><span class="p">(</span><span class="n">cnn</span><span class="p">,</span> <span class="n">normalization_mean</span><span class="p">,</span> <span class="n">normalization_std</span><span class="p">,</span>
<span class="n">content_img</span><span class="p">,</span> <span class="n">style_img</span><span class="p">,</span> <span class="n">input_img</span><span class="p">,</span> <span class="n">num_steps</span><span class="o">=</span><span class="mi">300</span><span class="p">,</span>
<span class="n">style_weight</span><span class="o">=</span><span class="mi">1000000</span><span class="p">,</span> <span class="n">content_weight</span><span class="o">=</span><span class="mi">1</span><span class="p">):</span>
<span class="w"> </span><span class="sd">"""Run the style transfer."""</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">'Building the style transfer model..'</span><span class="p">)</span>
<span class="n">model</span><span class="p">,</span> <span class="n">style_losses</span><span class="p">,</span> <span class="n">content_losses</span> <span class="o">=</span> <span class="n">get_style_model_and_losses</span><span class="p">(</span><span class="n">cnn</span><span class="p">,</span>
<span class="n">normalization_mean</span><span class="p">,</span> <span class="n">normalization_std</span><span class="p">,</span> <span class="n">style_img</span><span class="p">,</span> <span class="n">content_img</span><span class="p">)</span>
<span class="c1"># ๋ชจ๋ธ์ ๋งค๊ฐ๋ณ์๋ฅผ ์ ์ธํ ์
๋ ฅ์ ์ต์ ํํด์ผ ํ๋ฏ๋ก</span>
<span class="c1"># ์ด์ ๋ง์ถฐ์ requires_grad ๊ฐ์ ๊ฐฑ์ ํฉ๋๋ค.</span>
<span class="n">input_img</span><span class="o">.</span><span class="n">requires_grad_</span><span class="p">(</span><span class="kc">True</span><span class="p">)</span>
<span class="n">model</span><span class="o">.</span><span class="n">requires_grad_</span><span class="p">(</span><span class="kc">False</span><span class="p">)</span>
<span class="n">optimizer</span> <span class="o">=</span> <span class="n">get_input_optimizer</span><span class="p">(</span><span class="n">input_img</span><span class="p">)</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">'Optimizing..'</span><span class="p">)</span>
<span class="n">run</span> <span class="o">=</span> <span class="p">[</span><span class="mi">0</span><span class="p">]</span>
<span class="k">while</span> <span class="n">run</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o"><=</span> <span class="n">num_steps</span><span class="p">:</span>
<span class="k">def</span> <span class="nf">closure</span><span class="p">():</span>
<span class="c1"># ์
๋ฐ์ดํธ ๋ ์
๋ ฅ ์ด๋ฏธ์ง์ ๊ฐ์ ์์ </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">input_img</span><span class="o">.</span><span class="n">clamp_</span><span class="p">(</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">zero_grad</span><span class="p">()</span>
<span class="n">model</span><span class="p">(</span><span class="n">input_img</span><span class="p">)</span>
<span class="n">style_score</span> <span class="o">=</span> <span class="mi">0</span>
<span class="n">content_score</span> <span class="o">=</span> <span class="mi">0</span>
<span class="k">for</span> <span class="n">sl</span> <span class="ow">in</span> <span class="n">style_losses</span><span class="p">:</span>
<span class="n">style_score</span> <span class="o">+=</span> <span class="n">sl</span><span class="o">.</span><span class="n">loss</span>
<span class="k">for</span> <span class="n">cl</span> <span class="ow">in</span> <span class="n">content_losses</span><span class="p">:</span>
<span class="n">content_score</span> <span class="o">+=</span> <span class="n">cl</span><span class="o">.</span><span class="n">loss</span>
<span class="n">style_score</span> <span class="o">*=</span> <span class="n">style_weight</span>
<span class="n">content_score</span> <span class="o">*=</span> <span class="n">content_weight</span>
<span class="n">loss</span> <span class="o">=</span> <span class="n">style_score</span> <span class="o">+</span> <span class="n">content_score</span>
<span class="n">loss</span><span class="o">.</span><span class="n">backward</span><span class="p">()</span>
<span class="n">run</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">+=</span> <span class="mi">1</span>
<span class="k">if</span> <span class="n">run</span><span class="p">[</span><span class="mi">0</span><span class="p">]</span> <span class="o">%</span> <span class="mi">50</span> <span class="o">==</span> <span class="mi">0</span><span class="p">:</span>
<span class="nb">print</span><span class="p">(</span><span class="s2">"run </span><span class="si">{}</span><span class="s2">:"</span><span class="o">.</span><span class="n">format</span><span class="p">(</span><span class="n">run</span><span class="p">))</span>
<span class="nb">print</span><span class="p">(</span><span class="s1">'Style Loss : </span><span class="si">{:4f}</span><span class="s1"> Content Loss: </span><span class="si">{:4f}</span><span class="s1">'</span><span class="o">.</span><span class="n">format</span><span class="p">(</span>
<span class="n">style_score</span><span class="o">.</span><span class="n">item</span><span class="p">(),</span> <span class="n">content_score</span><span class="o">.</span><span class="n">item</span><span class="p">()))</span>
<span class="nb">print</span><span class="p">()</span>
<span class="k">return</span> <span class="n">style_score</span> <span class="o">+</span> <span class="n">content_score</span>
<span class="n">optimizer</span><span class="o">.</span><span class="n">step</span><span class="p">(</span><span class="n">closure</span><span class="p">)</span>
<span class="c1"># ๋ง์ง๋ง ์์ ...</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">input_img</span><span class="o">.</span><span class="n">clamp_</span><span class="p">(</span><span class="mi">0</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>
<span class="k">return</span> <span class="n">input_img</span>
</pre></div>
</div>
<p>๋ง์ง๋ง์ผ๋ก, ์๊ณ ๋ฆฌ์ฆ์ ์คํํฉ๋๋ค.</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">output</span> <span class="o">=</span> <span class="n">run_style_transfer</span><span class="p">(</span><span class="n">cnn</span><span class="p">,</span> <span class="n">cnn_normalization_mean</span><span class="p">,</span> <span class="n">cnn_normalization_std</span><span class="p">,</span>
<span class="n">content_img</span><span class="p">,</span> <span class="n">style_img</span><span class="p">,</span> <span class="n">input_img</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">imshow</span><span class="p">(</span><span class="n">output</span><span class="p">,</span> <span class="n">title</span><span class="o">=</span><span class="s1">'Output Image'</span><span class="p">)</span>
<span class="c1"># sphinx_gallery_thumbnail_number = 4</span>
<span class="n">plt</span><span class="o">.</span><span class="n">ioff</span><span class="p">()</span>
<span class="n">plt</span><span class="o">.</span><span class="n">show</span><span class="p">()</span>
</pre></div>
</div>
<img src="../_images/sphx_glr_neural_style_tutorial_004.png" srcset="../_images/sphx_glr_neural_style_tutorial_004.png" alt="Output Image" class = "sphx-glr-single-img"/><div class="sphx-glr-script-out highlight-none notranslate"><div class="highlight"><pre><span></span>Building the style transfer model..
Optimizing..
run [50]:
Style Loss : 4.081978 Content Loss: 4.146990
run [100]:
Style Loss : 1.158592 Content Loss: 3.051322
run [150]:
Style Loss : 0.726429 Content Loss: 2.666687
run [200]:
Style Loss : 0.490090 Content Loss: 2.498124
run [250]:
Style Loss : 0.350259 Content Loss: 2.407606
run [300]:
Style Loss : 0.265091 Content Loss: 2.350656
</pre></div>
</div>
<p class="sphx-glr-timing"><strong>Total running time of the script:</strong> ( 0 minutes 18.582 seconds)</p>
<div class="sphx-glr-footer sphx-glr-footer-example docutils container" id="sphx-glr-download-advanced-neural-style-tutorial-py">
<div class="sphx-glr-download sphx-glr-download-python docutils container">
<p><a class="reference download internal" download="" href="../_downloads/b3bcd49c4dd9be3a44c6c5d2e2a8b0cd/neural_style_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">neural_style_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/cad5020cab595c3bf83a518b7e4d4125/neural_style_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">neural_style_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>
<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 2018-2023, 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>
if((window.location.href.indexOf("/prototype/")!= -1) && (window.location.href.indexOf("/prototype/prototype_index")< 1))