Register
Login
Resources
Docs Blog Datasets Glossary Case Studies Tutorials & Webinars
Product
Data Engine LLMs Platform Enterprise
Pricing Explore
Connect to our Discord channel
Resources  
Docs Blog Datasets Glossary Case Studies Tutorials & Webinars
Explore Pricing
Product  
Data Engine LLMs Platform Enterprise
Connect to Dagshub Discord channel
Register
Login

Deci-AI
/
super-gradients
connected to https://github.com/Deci-AI/super-gradients.git

Watch 4
Issues 62 Pull Requests 44 Discussions 0

super_gradients.training.losses.html 121 KB
History Raw

You have to be logged in to leave a comment. Sign In
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
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069 
  1. <!DOCTYPE html>
    2. 

  2. <html class="writer-html5" lang="en" >
    3. 

  3. <head>
    4. 

  4.   <meta charset="utf-8" /><meta name="generator" content="Docutils 0.17.1: http://docutils.sourceforge.net/" />
    5. 

  5. 

  6.   <meta name="viewport" content="width=device-width, initial-scale=1.0" />
    7. 

  7.   <title>super_gradients.training.losses package &mdash; SuperGradients 1.0 documentation</title>
    8. 

  8.       <link rel="stylesheet" href="_static/pygments.css" type="text/css" />
    9. 

  9.       <link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
    10. 

  10.       <link rel="stylesheet" href="_static/graphviz.css" type="text/css" />
    11. 

  11.   <!--[if lt IE 9]>
    12. 

  12.     <script src="_static/js/html5shiv.min.js"></script>
    13. 

  13.   <![endif]-->
    14. 

  14.   
    15. 

  15.         <script data-url_root="./" id="documentation_options" src="_static/documentation_options.js"></script>
    16. 

  16.         <script src="_static/jquery.js"></script>
    17. 

  17.         <script src="_static/underscore.js"></script>
    18. 

  18.         <script src="_static/doctools.js"></script>
    19. 

  19.     <script src="_static/js/theme.js"></script>
    20. 

  20.     <link rel="index" title="Index" href="genindex.html" />
    21. 

  21.     <link rel="search" title="Search" href="search.html" /> 
    22. 

  22. </head>
    23. 

  23. 

  24. <body class="wy-body-for-nav"> 
    25. 

  25.   <div class="wy-grid-for-nav">
    26. 

  26.     <nav data-toggle="wy-nav-shift" class="wy-nav-side">
    27. 

  27.       <div class="wy-side-scroll">
    28. 

  28.         <div class="wy-side-nav-search" >
    29. 

  29.             <a href="index.html" class="icon icon-home"> SuperGradients
    30. 

  30.           </a>
    31. 

  31. <div role="search">
    32. 

  32.   <form id="rtd-search-form" class="wy-form" action="search.html" method="get">
    33. 

  33.     <input type="text" name="q" placeholder="Search docs" />
    34. 

  34.     <input type="hidden" name="check_keywords" value="yes" />
    35. 

  35.     <input type="hidden" name="area" value="default" />
    36. 

  36.   </form>
    37. 

  37. </div>
    38. 

  38.         </div><div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="Navigation menu">
    39. 

  39.               <p class="caption"><span class="caption-text">Welcome To SuperGradients</span></p>
    40. 

  40. <ul>
    41. 

  41. <li class="toctree-l1"><a class="reference internal" href="welcome.html">Fill our 4-question quick survey! We will raffle free SuperGradients swag between those who will participate -&gt; Fill Survey</a></li>
    42. 

  42. <li class="toctree-l1"><a class="reference internal" href="welcome.html#supergradients">SuperGradients</a></li>
    43. 

  43. </ul>
    44. 

  44. <p class="caption"><span class="caption-text">Technical Documentation</span></p>
    45. 

  45. <ul>
    46. 

  46. <li class="toctree-l1"><a class="reference internal" href="super_gradients.common.html">Common package</a></li>
    47. 

  47. <li class="toctree-l1"><a class="reference internal" href="super_gradients.training.html">Training package</a></li>
    48. 

  48. </ul>
    49. 

  49. <p class="caption"><span class="caption-text">User Guide</span></p>
    50. 

  50. <ul>
    51. 

  51. <li class="toctree-l1"><a class="reference internal" href="user_guide.html">What is SuperGradients?</a></li>
    52. 

  52. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#introducing-the-supergradients-library">Introducing the SuperGradients library</a></li>
    53. 

  53. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#installation">Installation</a></li>
    54. 

  54. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#integrating-your-training-code-complete-walkthrough">Integrating Your Training Code - Complete Walkthrough</a></li>
    55. 

  55. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#training-parameters">Training Parameters</a></li>
    56. 

  56. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#logs-and-checkpoints">Logs and Checkpoints</a></li>
    57. 

  57. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#dataset-parameters">Dataset Parameters</a></li>
    58. 

  58. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#network-architectures">Network Architectures</a></li>
    59. 

  59. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#pretrained-models">Pretrained Models</a></li>
    60. 

  60. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#how-to-reproduce-our-training-recipes">How To Reproduce Our Training Recipes</a></li>
    61. 

  61. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#professional-tools-integration">Professional Tools Integration</a></li>
    62. 

  62. <li class="toctree-l1"><a class="reference internal" href="user_guide.html#supergradients-faq">SuperGradients FAQ</a></li>
    63. 

  63. </ul>
    64. 

  64. 

  65.         </div>
    66. 

  66.       </div>
    67. 

  67.     </nav>
    68. 

  68. 

  69.     <section data-toggle="wy-nav-shift" class="wy-nav-content-wrap"><nav class="wy-nav-top" aria-label="Mobile navigation menu" >
    70. 

  70.           <i data-toggle="wy-nav-top" class="fa fa-bars"></i>
    71. 

  71.           <a href="index.html">SuperGradients</a>
    72. 

  72.       </nav>
    73. 

  73. 

  74.       <div class="wy-nav-content">
    75. 

  75.         <div class="rst-content">
    76. 

  76.           <div role="navigation" aria-label="Page navigation">
    77. 

  77.   <ul class="wy-breadcrumbs">
    78. 

  78.       <li><a href="index.html" class="icon icon-home"></a> &raquo;</li>
    79. 

  79.       <li>super_gradients.training.losses package</li>
    80. 

  80.       <li class="wy-breadcrumbs-aside">
    81. 

  81.             <a href="_sources/super_gradients.training.losses.rst.txt" rel="nofollow"> View page source</a>
    82. 

  82.       </li>
    83. 

  83.   </ul>
    84. 

  84.   <hr/>
    85. 

  85. </div>
    86. 

  86.           <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
    87. 

  87.            <div itemprop="articleBody">
    88. 

  88.              
    89. 

  89.   <section id="super-gradients-training-losses-package">
    90. 

  90. <h1>super_gradients.training.losses package<a class="headerlink" href="#super-gradients-training-losses-package" title="Permalink to this headline"></a></h1>
    91. 

  91. <section id="submodules">
    92. 

  92. <h2>Submodules<a class="headerlink" href="#submodules" title="Permalink to this headline"></a></h2>
    93. 

  93. </section>
    94. 

  94. <section id="module-super_gradients.training.losses.all_losses">
    95. 

  95. <span id="super-gradients-training-losses-all-losses-module"></span><h2>super_gradients.training.losses.all_losses module<a class="headerlink" href="#module-super_gradients.training.losses.all_losses" title="Permalink to this headline"></a></h2>
    96. 

  96. </section>
    97. 

  97. <section id="module-super_gradients.training.losses.ddrnet_loss">
    98. 

  98. <span id="super-gradients-training-losses-ddrnet-loss-module"></span><h2>super_gradients.training.losses.ddrnet_loss module<a class="headerlink" href="#module-super_gradients.training.losses.ddrnet_loss" title="Permalink to this headline"></a></h2>
    99. 

  99. <dl class="py class">
    100. 

  100. <dt class="sig sig-object py" id="super_gradients.training.losses.ddrnet_loss.DDRNetLoss">
    101. 

  101. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.ddrnet_loss.</span></span><span class="sig-name descname"><span class="pre">DDRNetLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">threshold</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.7</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ohem_percentage</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.1</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">weights</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">list</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">[1.0,</span> <span class="pre">0.4]</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_label</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">255</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_pixels_exclude_ignored</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">False</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ddrnet_loss.html#DDRNetLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ddrnet_loss.DDRNetLoss" title="Permalink to this definition"></a></dt>
    102. 

  102. <dd><p>Bases: <a class="reference internal" href="#super_gradients.training.losses.ohem_ce_loss.OhemCELoss" title="super_gradients.training.losses.ohem_ce_loss.OhemCELoss"><code class="xref py py-class docutils literal notranslate"><span class="pre">super_gradients.training.losses.ohem_ce_loss.OhemCELoss</span></code></a></p>
    103. 

  103. <dl class="py method">
    104. 

  104. <dt class="sig sig-object py" id="super_gradients.training.losses.ddrnet_loss.DDRNetLoss.forward">
    105. 

  105. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">predictions_list</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">list</span><span class="p"><span class="pre">,</span> </span><span class="pre">tuple</span><span class="p"><span class="pre">,</span> </span><span class="pre">torch.Tensor</span><span class="p"><span class="pre">]</span></span></span></em>, <em class="sig-param"><span class="n"><span class="pre">targets</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.Tensor</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ddrnet_loss.html#DDRNetLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ddrnet_loss.DDRNetLoss.forward" title="Permalink to this definition"></a></dt>
    106. 

  106. <dd><p>Defines the computation performed at every call.</p>
    107. 

  107. <p>Should be overridden by all subclasses.</p>
    108. 

  108. <div class="admonition note">
    109. 

  109. <p class="admonition-title">Note</p>
    110. 

  110. <p>Although the recipe for forward pass needs to be defined within
    111. 

  111. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    112. 

  112. instead of this since the former takes care of running the
    113. 

  113. registered hooks while the latter silently ignores them.</p>
    114. 

  114. </div>
    115. 

  115. </dd></dl>
    116. 

  116. 

  117. <dl class="py attribute">
    118. 

  118. <dt class="sig sig-object py" id="super_gradients.training.losses.ddrnet_loss.DDRNetLoss.reduction">
    119. 

  119. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.ddrnet_loss.DDRNetLoss.reduction" title="Permalink to this definition"></a></dt>
    120. 

  120. <dd></dd></dl>
    121. 

  121. 

  122. </dd></dl>
    123. 

  123. 

  124. </section>
    125. 

  125. <section id="module-super_gradients.training.losses.focal_loss">
    126. 

  126. <span id="super-gradients-training-losses-focal-loss-module"></span><h2>super_gradients.training.losses.focal_loss module<a class="headerlink" href="#module-super_gradients.training.losses.focal_loss" title="Permalink to this headline"></a></h2>
    127. 

  127. <dl class="py class">
    128. 

  128. <dt class="sig sig-object py" id="super_gradients.training.losses.focal_loss.FocalLoss">
    129. 

  129. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.focal_loss.</span></span><span class="sig-name descname"><span class="pre">FocalLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">loss_fcn</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.nn.modules.loss.BCEWithLogitsLoss</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">gamma</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">1.5</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">alpha</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.25</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/focal_loss.html#FocalLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.focal_loss.FocalLoss" title="Permalink to this definition"></a></dt>
    130. 

  130. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    131. 

  131. <p>Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)</p>
    132. 

  132. <dl class="py attribute">
    133. 

  133. <dt class="sig sig-object py" id="super_gradients.training.losses.focal_loss.FocalLoss.reduction">
    134. 

  134. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.focal_loss.FocalLoss.reduction" title="Permalink to this definition"></a></dt>
    135. 

  135. <dd></dd></dl>
    136. 

  136. 

  137. <dl class="py method">
    138. 

  138. <dt class="sig sig-object py" id="super_gradients.training.losses.focal_loss.FocalLoss.forward">
    139. 

  139. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">pred</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">true</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/focal_loss.html#FocalLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.focal_loss.FocalLoss.forward" title="Permalink to this definition"></a></dt>
    140. 

  140. <dd><p>Defines the computation performed at every call.</p>
    141. 

  141. <p>Should be overridden by all subclasses.</p>
    142. 

  142. <div class="admonition note">
    143. 

  143. <p class="admonition-title">Note</p>
    144. 

  144. <p>Although the recipe for forward pass needs to be defined within
    145. 

  145. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    146. 

  146. instead of this since the former takes care of running the
    147. 

  147. registered hooks while the latter silently ignores them.</p>
    148. 

  148. </div>
    149. 

  149. </dd></dl>
    150. 

  150. 

  151. </dd></dl>
    152. 

  152. 

  153. </section>
    154. 

  154. <section id="module-super_gradients.training.losses.label_smoothing_cross_entropy_loss">
    155. 

  155. <span id="super-gradients-training-losses-label-smoothing-cross-entropy-loss-module"></span><h2>super_gradients.training.losses.label_smoothing_cross_entropy_loss module<a class="headerlink" href="#module-super_gradients.training.losses.label_smoothing_cross_entropy_loss" title="Permalink to this headline"></a></h2>
    156. 

  156. <dl class="py function">
    157. 

  157. <dt class="sig sig-object py" id="super_gradients.training.losses.label_smoothing_cross_entropy_loss.onehot">
    158. 

  158. <span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.label_smoothing_cross_entropy_loss.</span></span><span class="sig-name descname"><span class="pre">onehot</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">indexes</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">N</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_index</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/label_smoothing_cross_entropy_loss.html#onehot"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.label_smoothing_cross_entropy_loss.onehot" title="Permalink to this definition"></a></dt>
    159. 

  159. <dd><p>Creates a one-hot representation of indexes with N possible entries
    160. 

  160. if N is not specified, it will suit the maximum index appearing.
    161. 

  161. indexes is a long-tensor of indexes
    162. 

  162. ignore_index will be zero in onehot representation</p>
    163. 

  163. </dd></dl>
    164. 

  164. 

  165. <dl class="py function">
    166. 

  166. <dt class="sig sig-object py" id="super_gradients.training.losses.label_smoothing_cross_entropy_loss.cross_entropy">
    167. 

  167. <span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.label_smoothing_cross_entropy_loss.</span></span><span class="sig-name descname"><span class="pre">cross_entropy</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">inputs</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">target</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">weight</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_index</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">-</span> <span class="pre">100</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">reduction</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">'mean'</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">smooth_eps</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">smooth_dist</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">from_logits</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/label_smoothing_cross_entropy_loss.html#cross_entropy"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.label_smoothing_cross_entropy_loss.cross_entropy" title="Permalink to this definition"></a></dt>
    168. 

  168. <dd><p>cross entropy loss, with support for target distributions and label smoothing <a class="reference external" href="https://arxiv.org/abs/1512.00567">https://arxiv.org/abs/1512.00567</a></p>
    169. 

  169. </dd></dl>
    170. 

  170. 

  171. <dl class="py class">
    172. 

  172. <dt class="sig sig-object py" id="super_gradients.training.losses.label_smoothing_cross_entropy_loss.LabelSmoothingCrossEntropyLoss">
    173. 

  173. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.label_smoothing_cross_entropy_loss.</span></span><span class="sig-name descname"><span class="pre">LabelSmoothingCrossEntropyLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">weight</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_index</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">-</span> <span class="pre">100</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">reduction</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">'mean'</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">smooth_eps</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">smooth_dist</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">from_logits</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/label_smoothing_cross_entropy_loss.html#LabelSmoothingCrossEntropyLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.label_smoothing_cross_entropy_loss.LabelSmoothingCrossEntropyLoss" title="Permalink to this definition"></a></dt>
    174. 

  174. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss.CrossEntropyLoss</span></code></p>
    175. 

  175. <p>CrossEntropyLoss - with ability to recieve distrbution as targets, and optional label smoothing</p>
    176. 

  176. <dl class="py method">
    177. 

  177. <dt class="sig sig-object py" id="super_gradients.training.losses.label_smoothing_cross_entropy_loss.LabelSmoothingCrossEntropyLoss.forward">
    178. 

  178. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">input</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">target</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">smooth_dist</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/label_smoothing_cross_entropy_loss.html#LabelSmoothingCrossEntropyLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.label_smoothing_cross_entropy_loss.LabelSmoothingCrossEntropyLoss.forward" title="Permalink to this definition"></a></dt>
    179. 

  179. <dd><p>Defines the computation performed at every call.</p>
    180. 

  180. <p>Should be overridden by all subclasses.</p>
    181. 

  181. <div class="admonition note">
    182. 

  182. <p class="admonition-title">Note</p>
    183. 

  183. <p>Although the recipe for forward pass needs to be defined within
    184. 

  184. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    185. 

  185. instead of this since the former takes care of running the
    186. 

  186. registered hooks while the latter silently ignores them.</p>
    187. 

  187. </div>
    188. 

  188. </dd></dl>
    189. 

  189. 

  190. <dl class="py attribute">
    191. 

  191. <dt class="sig sig-object py" id="super_gradients.training.losses.label_smoothing_cross_entropy_loss.LabelSmoothingCrossEntropyLoss.ignore_index">
    192. 

  192. <span class="sig-name descname"><span class="pre">ignore_index</span></span><em class="property"><span class="pre">:</span> <span class="pre">int</span></em><a class="headerlink" href="#super_gradients.training.losses.label_smoothing_cross_entropy_loss.LabelSmoothingCrossEntropyLoss.ignore_index" title="Permalink to this definition"></a></dt>
    193. 

  193. <dd></dd></dl>
    194. 

  194. 

  195. <dl class="py attribute">
    196. 

  196. <dt class="sig sig-object py" id="super_gradients.training.losses.label_smoothing_cross_entropy_loss.LabelSmoothingCrossEntropyLoss.label_smoothing">
    197. 

  197. <span class="sig-name descname"><span class="pre">label_smoothing</span></span><em class="property"><span class="pre">:</span> <span class="pre">float</span></em><a class="headerlink" href="#super_gradients.training.losses.label_smoothing_cross_entropy_loss.LabelSmoothingCrossEntropyLoss.label_smoothing" title="Permalink to this definition"></a></dt>
    198. 

  198. <dd></dd></dl>
    199. 

  199. 

  200. </dd></dl>
    201. 

  201. 

  202. </section>
    203. 

  203. <section id="module-super_gradients.training.losses.ohem_ce_loss">
    204. 

  204. <span id="super-gradients-training-losses-ohem-ce-loss-module"></span><h2>super_gradients.training.losses.ohem_ce_loss module<a class="headerlink" href="#module-super_gradients.training.losses.ohem_ce_loss" title="Permalink to this headline"></a></h2>
    205. 

  205. <dl class="py class">
    206. 

  206. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemLoss">
    207. 

  207. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.ohem_ce_loss.</span></span><span class="sig-name descname"><span class="pre">OhemLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">threshold</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">mining_percent</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.1</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_lb</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">-</span> <span class="pre">100</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_pixels_exclude_ignored</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">True</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">criteria</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.nn.modules.loss._Loss</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ohem_ce_loss.html#OhemLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemLoss" title="Permalink to this definition"></a></dt>
    208. 

  208. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    209. 

  209. <p>OhemLoss - Online Hard Example Mining Cross Entropy Loss</p>
    210. 

  210. <dl class="py method">
    211. 

  211. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemLoss.forward">
    212. 

  212. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">logits</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">labels</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ohem_ce_loss.html#OhemLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemLoss.forward" title="Permalink to this definition"></a></dt>
    213. 

  213. <dd><p>Defines the computation performed at every call.</p>
    214. 

  214. <p>Should be overridden by all subclasses.</p>
    215. 

  215. <div class="admonition note">
    216. 

  216. <p class="admonition-title">Note</p>
    217. 

  217. <p>Although the recipe for forward pass needs to be defined within
    218. 

  218. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    219. 

  219. instead of this since the former takes care of running the
    220. 

  220. registered hooks while the latter silently ignores them.</p>
    221. 

  221. </div>
    222. 

  222. </dd></dl>
    223. 

  223. 

  224. <dl class="py attribute">
    225. 

  225. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemLoss.reduction">
    226. 

  226. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemLoss.reduction" title="Permalink to this definition"></a></dt>
    227. 

  227. <dd></dd></dl>
    228. 

  228. 

  229. </dd></dl>
    230. 

  230. 

  231. <dl class="py class">
    232. 

  232. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemCELoss">
    233. 

  233. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.ohem_ce_loss.</span></span><span class="sig-name descname"><span class="pre">OhemCELoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">threshold</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">mining_percent</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.1</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_lb</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">-</span> <span class="pre">100</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_pixels_exclude_ignored</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ohem_ce_loss.html#OhemCELoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemCELoss" title="Permalink to this definition"></a></dt>
    234. 

  234. <dd><p>Bases: <a class="reference internal" href="#super_gradients.training.losses.ohem_ce_loss.OhemLoss" title="super_gradients.training.losses.ohem_ce_loss.OhemLoss"><code class="xref py py-class docutils literal notranslate"><span class="pre">super_gradients.training.losses.ohem_ce_loss.OhemLoss</span></code></a></p>
    235. 

  235. <p>OhemLoss - Online Hard Example Mining Cross Entropy Loss</p>
    236. 

  236. <dl class="py attribute">
    237. 

  237. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemCELoss.reduction">
    238. 

  238. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemCELoss.reduction" title="Permalink to this definition"></a></dt>
    239. 

  239. <dd></dd></dl>
    240. 

  240. 

  241. <dl class="py attribute">
    242. 

  242. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemCELoss.training">
    243. 

  243. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemCELoss.training" title="Permalink to this definition"></a></dt>
    244. 

  244. <dd></dd></dl>
    245. 

  245. 

  246. </dd></dl>
    247. 

  247. 

  248. <dl class="py class">
    249. 

  249. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemBCELoss">
    250. 

  250. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.ohem_ce_loss.</span></span><span class="sig-name descname"><span class="pre">OhemBCELoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">threshold</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">mining_percent</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.1</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_lb</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">-</span> <span class="pre">100</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_pixels_exclude_ignored</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ohem_ce_loss.html#OhemBCELoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemBCELoss" title="Permalink to this definition"></a></dt>
    251. 

  251. <dd><p>Bases: <a class="reference internal" href="#super_gradients.training.losses.ohem_ce_loss.OhemLoss" title="super_gradients.training.losses.ohem_ce_loss.OhemLoss"><code class="xref py py-class docutils literal notranslate"><span class="pre">super_gradients.training.losses.ohem_ce_loss.OhemLoss</span></code></a></p>
    252. 

  252. <p>OhemBCELoss - Online Hard Example Mining Binary Cross Entropy Loss</p>
    253. 

  253. <dl class="py method">
    254. 

  254. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemBCELoss.forward">
    255. 

  255. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">logits</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">labels</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ohem_ce_loss.html#OhemBCELoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemBCELoss.forward" title="Permalink to this definition"></a></dt>
    256. 

  256. <dd><p>Defines the computation performed at every call.</p>
    257. 

  257. <p>Should be overridden by all subclasses.</p>
    258. 

  258. <div class="admonition note">
    259. 

  259. <p class="admonition-title">Note</p>
    260. 

  260. <p>Although the recipe for forward pass needs to be defined within
    261. 

  261. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    262. 

  262. instead of this since the former takes care of running the
    263. 

  263. registered hooks while the latter silently ignores them.</p>
    264. 

  264. </div>
    265. 

  265. </dd></dl>
    266. 

  266. 

  267. <dl class="py attribute">
    268. 

  268. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemBCELoss.reduction">
    269. 

  269. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemBCELoss.reduction" title="Permalink to this definition"></a></dt>
    270. 

  270. <dd></dd></dl>
    271. 

  271. 

  272. <dl class="py attribute">
    273. 

  273. <dt class="sig sig-object py" id="super_gradients.training.losses.ohem_ce_loss.OhemBCELoss.training">
    274. 

  274. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.ohem_ce_loss.OhemBCELoss.training" title="Permalink to this definition"></a></dt>
    275. 

  275. <dd></dd></dl>
    276. 

  276. 

  277. </dd></dl>
    278. 

  278. 

  279. </section>
    280. 

  280. <section id="module-super_gradients.training.losses.r_squared_loss">
    281. 

  281. <span id="super-gradients-training-losses-r-squared-loss-module"></span><h2>super_gradients.training.losses.r_squared_loss module<a class="headerlink" href="#module-super_gradients.training.losses.r_squared_loss" title="Permalink to this headline"></a></h2>
    282. 

  282. <dl class="py class">
    283. 

  283. <dt class="sig sig-object py" id="super_gradients.training.losses.r_squared_loss.RSquaredLoss">
    284. 

  284. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.r_squared_loss.</span></span><span class="sig-name descname"><span class="pre">RSquaredLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">size_average</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">reduce</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">reduction</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">str</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">'mean'</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/r_squared_loss.html#RSquaredLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.r_squared_loss.RSquaredLoss" title="Permalink to this definition"></a></dt>
    285. 

  285. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    286. 

  286. <dl class="py method">
    287. 

  287. <dt class="sig sig-object py" id="super_gradients.training.losses.r_squared_loss.RSquaredLoss.forward">
    288. 

  288. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">output</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">target</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/r_squared_loss.html#RSquaredLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.r_squared_loss.RSquaredLoss.forward" title="Permalink to this definition"></a></dt>
    289. 

  289. <dd><p>Computes the R-squared for the output and target values
    290. 

  290. :param output: Tensor / Numpy / List</p>
    291. 

  291. <blockquote>
    292. 

  292. <div><p>The prediction</p>
    293. 

  293. </div></blockquote>
    294. 

  294. <dl class="field-list simple">
    295. 

  295. <dt class="field-odd">Parameters</dt>
    296. 

  296. <dd class="field-odd"><p><strong>target</strong> – Tensor / Numpy / List
    297. 

  297. The corresponding lables</p>
    298. 

  298. </dd>
    299. 

  299. </dl>
    300. 

  300. </dd></dl>
    301. 

  301. 

  302. <dl class="py attribute">
    303. 

  303. <dt class="sig sig-object py" id="super_gradients.training.losses.r_squared_loss.RSquaredLoss.reduction">
    304. 

  304. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.r_squared_loss.RSquaredLoss.reduction" title="Permalink to this definition"></a></dt>
    305. 

  305. <dd></dd></dl>
    306. 

  306. 

  307. </dd></dl>
    308. 

  308. 

  309. </section>
    310. 

  310. <section id="module-super_gradients.training.losses.shelfnet_ohem_loss">
    311. 

  311. <span id="super-gradients-training-losses-shelfnet-ohem-loss-module"></span><h2>super_gradients.training.losses.shelfnet_ohem_loss module<a class="headerlink" href="#module-super_gradients.training.losses.shelfnet_ohem_loss" title="Permalink to this headline"></a></h2>
    312. 

  312. <dl class="py class">
    313. 

  313. <dt class="sig sig-object py" id="super_gradients.training.losses.shelfnet_ohem_loss.ShelfNetOHEMLoss">
    314. 

  314. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.shelfnet_ohem_loss.</span></span><span class="sig-name descname"><span class="pre">ShelfNetOHEMLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">threshold</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.7</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">mining_percent</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.0001</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_lb</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">255</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/shelfnet_ohem_loss.html#ShelfNetOHEMLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.shelfnet_ohem_loss.ShelfNetOHEMLoss" title="Permalink to this definition"></a></dt>
    315. 

  315. <dd><p>Bases: <a class="reference internal" href="#super_gradients.training.losses.ohem_ce_loss.OhemCELoss" title="super_gradients.training.losses.ohem_ce_loss.OhemCELoss"><code class="xref py py-class docutils literal notranslate"><span class="pre">super_gradients.training.losses.ohem_ce_loss.OhemCELoss</span></code></a></p>
    316. 

  316. <dl class="py method">
    317. 

  317. <dt class="sig sig-object py" id="super_gradients.training.losses.shelfnet_ohem_loss.ShelfNetOHEMLoss.forward">
    318. 

  318. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">predictions_list</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">list</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">targets</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/shelfnet_ohem_loss.html#ShelfNetOHEMLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.shelfnet_ohem_loss.ShelfNetOHEMLoss.forward" title="Permalink to this definition"></a></dt>
    319. 

  319. <dd><p>Defines the computation performed at every call.</p>
    320. 

  320. <p>Should be overridden by all subclasses.</p>
    321. 

  321. <div class="admonition note">
    322. 

  322. <p class="admonition-title">Note</p>
    323. 

  323. <p>Although the recipe for forward pass needs to be defined within
    324. 

  324. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    325. 

  325. instead of this since the former takes care of running the
    326. 

  326. registered hooks while the latter silently ignores them.</p>
    327. 

  327. </div>
    328. 

  328. </dd></dl>
    329. 

  329. 

  330. <dl class="py attribute">
    331. 

  331. <dt class="sig sig-object py" id="super_gradients.training.losses.shelfnet_ohem_loss.ShelfNetOHEMLoss.reduction">
    332. 

  332. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.shelfnet_ohem_loss.ShelfNetOHEMLoss.reduction" title="Permalink to this definition"></a></dt>
    333. 

  333. <dd></dd></dl>
    334. 

  334. 

  335. <dl class="py attribute">
    336. 

  336. <dt class="sig sig-object py" id="super_gradients.training.losses.shelfnet_ohem_loss.ShelfNetOHEMLoss.training">
    337. 

  337. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.shelfnet_ohem_loss.ShelfNetOHEMLoss.training" title="Permalink to this definition"></a></dt>
    338. 

  338. <dd></dd></dl>
    339. 

  339. 

  340. </dd></dl>
    341. 

  341. 

  342. </section>
    343. 

  343. <section id="module-super_gradients.training.losses.shelfnet_semantic_encoding_loss">
    344. 

  344. <span id="super-gradients-training-losses-shelfnet-semantic-encoding-loss-module"></span><h2>super_gradients.training.losses.shelfnet_semantic_encoding_loss module<a class="headerlink" href="#module-super_gradients.training.losses.shelfnet_semantic_encoding_loss" title="Permalink to this headline"></a></h2>
    345. 

  345. <dl class="py class">
    346. 

  346. <dt class="sig sig-object py" id="super_gradients.training.losses.shelfnet_semantic_encoding_loss.ShelfNetSemanticEncodingLoss">
    347. 

  347. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.shelfnet_semantic_encoding_loss.</span></span><span class="sig-name descname"><span class="pre">ShelfNetSemanticEncodingLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">se_weight</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.2</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">nclass</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">21</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">aux_weight</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.4</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">weight</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_index</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">-</span> <span class="pre">1</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/shelfnet_semantic_encoding_loss.html#ShelfNetSemanticEncodingLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.shelfnet_semantic_encoding_loss.ShelfNetSemanticEncodingLoss" title="Permalink to this definition"></a></dt>
    348. 

  348. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss.CrossEntropyLoss</span></code></p>
    349. 

  349. <p>2D Cross Entropy Loss with Auxilary Loss</p>
    350. 

  350. <dl class="py method">
    351. 

  351. <dt class="sig sig-object py" id="super_gradients.training.losses.shelfnet_semantic_encoding_loss.ShelfNetSemanticEncodingLoss.forward">
    352. 

  352. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">logits</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">labels</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/shelfnet_semantic_encoding_loss.html#ShelfNetSemanticEncodingLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.shelfnet_semantic_encoding_loss.ShelfNetSemanticEncodingLoss.forward" title="Permalink to this definition"></a></dt>
    353. 

  353. <dd><p>Defines the computation performed at every call.</p>
    354. 

  354. <p>Should be overridden by all subclasses.</p>
    355. 

  355. <div class="admonition note">
    356. 

  356. <p class="admonition-title">Note</p>
    357. 

  357. <p>Although the recipe for forward pass needs to be defined within
    358. 

  358. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    359. 

  359. instead of this since the former takes care of running the
    360. 

  360. registered hooks while the latter silently ignores them.</p>
    361. 

  361. </div>
    362. 

  362. </dd></dl>
    363. 

  363. 

  364. <dl class="py attribute">
    365. 

  365. <dt class="sig sig-object py" id="super_gradients.training.losses.shelfnet_semantic_encoding_loss.ShelfNetSemanticEncodingLoss.ignore_index">
    366. 

  366. <span class="sig-name descname"><span class="pre">ignore_index</span></span><em class="property"><span class="pre">:</span> <span class="pre">int</span></em><a class="headerlink" href="#super_gradients.training.losses.shelfnet_semantic_encoding_loss.ShelfNetSemanticEncodingLoss.ignore_index" title="Permalink to this definition"></a></dt>
    367. 

  367. <dd></dd></dl>
    368. 

  368. 

  369. <dl class="py attribute">
    370. 

  370. <dt class="sig sig-object py" id="super_gradients.training.losses.shelfnet_semantic_encoding_loss.ShelfNetSemanticEncodingLoss.label_smoothing">
    371. 

  371. <span class="sig-name descname"><span class="pre">label_smoothing</span></span><em class="property"><span class="pre">:</span> <span class="pre">float</span></em><a class="headerlink" href="#super_gradients.training.losses.shelfnet_semantic_encoding_loss.ShelfNetSemanticEncodingLoss.label_smoothing" title="Permalink to this definition"></a></dt>
    372. 

  372. <dd></dd></dl>
    373. 

  373. 

  374. </dd></dl>
    375. 

  375. 

  376. </section>
    377. 

  377. <section id="module-super_gradients.training.losses.ssd_loss">
    378. 

  378. <span id="super-gradients-training-losses-ssd-loss-module"></span><h2>super_gradients.training.losses.ssd_loss module<a class="headerlink" href="#module-super_gradients.training.losses.ssd_loss" title="Permalink to this headline"></a></h2>
    379. 

  379. <dl class="py class">
    380. 

  380. <dt class="sig sig-object py" id="super_gradients.training.losses.ssd_loss.HardMiningCrossEntropyLoss">
    381. 

  381. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.ssd_loss.</span></span><span class="sig-name descname"><span class="pre">HardMiningCrossEntropyLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">neg_pos_ratio</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ssd_loss.html#HardMiningCrossEntropyLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ssd_loss.HardMiningCrossEntropyLoss" title="Permalink to this definition"></a></dt>
    382. 

  382. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    383. 

  383. <p>L_cls = [CE of all positives] + [CE of the hardest backgrounds]
    384. 

  384. where the second term is built from [neg_pos_ratio * positive pairs] background cells with the highest CE
    385. 

  385. (the hardest background cells)</p>
    386. 

  386. <dl class="py method">
    387. 

  387. <dt class="sig sig-object py" id="super_gradients.training.losses.ssd_loss.HardMiningCrossEntropyLoss.forward">
    388. 

  388. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">pred_labels</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">target_labels</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ssd_loss.html#HardMiningCrossEntropyLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ssd_loss.HardMiningCrossEntropyLoss.forward" title="Permalink to this definition"></a></dt>
    389. 

  389. <dd><p>Defines the computation performed at every call.</p>
    390. 

  390. <p>Should be overridden by all subclasses.</p>
    391. 

  391. <div class="admonition note">
    392. 

  392. <p class="admonition-title">Note</p>
    393. 

  393. <p>Although the recipe for forward pass needs to be defined within
    394. 

  394. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    395. 

  395. instead of this since the former takes care of running the
    396. 

  396. registered hooks while the latter silently ignores them.</p>
    397. 

  397. </div>
    398. 

  398. </dd></dl>
    399. 

  399. 

  400. <dl class="py attribute">
    401. 

  401. <dt class="sig sig-object py" id="super_gradients.training.losses.ssd_loss.HardMiningCrossEntropyLoss.reduction">
    402. 

  402. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.ssd_loss.HardMiningCrossEntropyLoss.reduction" title="Permalink to this definition"></a></dt>
    403. 

  403. <dd></dd></dl>
    404. 

  404. 

  405. </dd></dl>
    406. 

  406. 

  407. <dl class="py class">
    408. 

  408. <dt class="sig sig-object py" id="super_gradients.training.losses.ssd_loss.SSDLoss">
    409. 

  409. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.ssd_loss.</span></span><span class="sig-name descname"><span class="pre">SSDLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">dboxes</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><a class="reference internal" href="super_gradients.training.utils.html#super_gradients.training.utils.ssd_utils.DefaultBoxes" title="super_gradients.training.utils.ssd_utils.DefaultBoxes"><span class="pre">super_gradients.training.utils.ssd_utils.DefaultBoxes</span></a></span></em>, <em class="sig-param"><span class="n"><span class="pre">alpha</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">1.0</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">iou_thresh</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.5</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">neg_pos_ratio</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">3.0</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ssd_loss.html#SSDLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ssd_loss.SSDLoss" title="Permalink to this definition"></a></dt>
    410. 

  410. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    411. 

  411. <blockquote>
    412. 

  412. <div><p>Implements the loss as the sum of the followings:
    413. 

  413. 1. Confidence Loss: All labels, with hard negative mining
    414. 

  414. 2. Localization Loss: Only on positive labels</p>
    415. 

  415. </div></blockquote>
    416. 

  416. <dl class="simple">
    417. 

  417. <dt>L = (2 - alpha) * L_l1 + alpha * L_cls, where</dt><dd><ul class="simple">
    418. 

  418. <li><p>L_cls is HardMiningCrossEntropyLoss</p></li>
    419. 

  419. <li><p>L_l1 = [SmoothL1Loss for all positives]</p></li>
    420. 

  420. </ul>
    421. 

  421. </dd>
    422. 

  422. </dl>
    423. 

  423. <dl class="py method">
    424. 

  424. <dt class="sig sig-object py" id="super_gradients.training.losses.ssd_loss.SSDLoss.match_dboxes">
    425. 

  425. <span class="sig-name descname"><span class="pre">match_dboxes</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">targets</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ssd_loss.html#SSDLoss.match_dboxes"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ssd_loss.SSDLoss.match_dboxes" title="Permalink to this definition"></a></dt>
    426. 

  426. <dd><p>creates tensors with target boxes and labels for each dboxes, so with the same len as dboxes.</p>
    427. 

  427. <ul class="simple">
    428. 

  428. <li><p>Each GT is assigned with a grid cell with the highest IoU, this creates a pair for each GT and some cells;</p></li>
    429. 

  429. <li><p>The rest of grid cells are assigned to a GT with the highest IoU, assuming it’s &gt; self.iou_thresh;
    430. 

  430. If this condition is not met the grid cell is marked as background</p></li>
    431. 

  431. </ul>
    432. 

  432. <p>GT-wise: one to many
    433. 

  433. Grid-cell-wise: one to one</p>
    434. 

  434. <dl class="field-list simple">
    435. 

  435. <dt class="field-odd">Parameters</dt>
    436. 

  436. <dd class="field-odd"><p><strong>targets</strong> – a tensor containing the boxes for a single image;
    437. 

  437. shape [num_boxes, 6] (image_id, label, x, y, w, h)</p>
    438. 

  438. </dd>
    439. 

  439. <dt class="field-even">Returns</dt>
    440. 

  440. <dd class="field-even"><p>two tensors
    441. 

  441. boxes - shape of dboxes [4, num_dboxes] (x,y,w,h)
    442. 

  442. labels - sahpe [num_dboxes]</p>
    443. 

  443. </dd>
    444. 

  444. </dl>
    445. 

  445. </dd></dl>
    446. 

  446. 

  447. <dl class="py method">
    448. 

  448. <dt class="sig sig-object py" id="super_gradients.training.losses.ssd_loss.SSDLoss.forward">
    449. 

  449. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">predictions</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Tuple</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">targets</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ssd_loss.html#SSDLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ssd_loss.SSDLoss.forward" title="Permalink to this definition"></a></dt>
    450. 

  450. <dd><dl class="simple">
    451. 

  451. <dt>Compute the loss</dt><dd><p>:param predictions - predictions tensor coming from the network,
    452. 

  452. tuple with shapes ([Batch Size, 4, num_dboxes], [Batch Size, num_classes + 1, num_dboxes])
    453. 

  453. were predictions have logprobs for background and other classes
    454. 

  454. :param targets - targets for the batch. [num targets, 6] (index in batch, label, x,y,w,h)</p>
    455. 

  455. </dd>
    456. 

  456. </dl>
    457. 

  457. </dd></dl>
    458. 

  458. 

  459. <dl class="py attribute">
    460. 

  460. <dt class="sig sig-object py" id="super_gradients.training.losses.ssd_loss.SSDLoss.reduction">
    461. 

  461. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.ssd_loss.SSDLoss.reduction" title="Permalink to this definition"></a></dt>
    462. 

  462. <dd></dd></dl>
    463. 

  463. 

  464. </dd></dl>
    465. 

  465. 

  466. </section>
    467. 

  467. <section id="super-gradients-training-losses-yolo-v3-loss-module">
    468. 

  468. <h2>super_gradients.training.losses.yolo_v3_loss module<a class="headerlink" href="#super-gradients-training-losses-yolo-v3-loss-module" title="Permalink to this headline"></a></h2>
    469. 

  469. </section>
    470. 

  470. <section id="super-gradients-training-losses-yolo-v5-loss-module">
    471. 

  471. <h2>super_gradients.training.losses.yolo_v5_loss module<a class="headerlink" href="#super-gradients-training-losses-yolo-v5-loss-module" title="Permalink to this headline"></a></h2>
    472. 

  472. </section>
    473. 

  473. <section id="module-super_gradients.training.losses">
    474. 

  474. <span id="module-contents"></span><h2>Module contents<a class="headerlink" href="#module-super_gradients.training.losses" title="Permalink to this headline"></a></h2>
    475. 

  475. <dl class="py class">
    476. 

  476. <dt class="sig sig-object py" id="super_gradients.training.losses.FocalLoss">
    477. 

  477. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">FocalLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">loss_fcn</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.nn.modules.loss.BCEWithLogitsLoss</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">gamma</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">1.5</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">alpha</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.25</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/focal_loss.html#FocalLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.FocalLoss" title="Permalink to this definition"></a></dt>
    478. 

  478. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    479. 

  479. <p>Wraps focal loss around existing loss_fcn(), i.e. criteria = FocalLoss(nn.BCEWithLogitsLoss(), gamma=1.5)</p>
    480. 

  480. <dl class="py attribute">
    481. 

  481. <dt class="sig sig-object py" id="super_gradients.training.losses.FocalLoss.reduction">
    482. 

  482. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.FocalLoss.reduction" title="Permalink to this definition"></a></dt>
    483. 

  483. <dd></dd></dl>
    484. 

  484. 

  485. <dl class="py method">
    486. 

  486. <dt class="sig sig-object py" id="super_gradients.training.losses.FocalLoss.forward">
    487. 

  487. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">pred</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">true</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/focal_loss.html#FocalLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.FocalLoss.forward" title="Permalink to this definition"></a></dt>
    488. 

  488. <dd><p>Defines the computation performed at every call.</p>
    489. 

  489. <p>Should be overridden by all subclasses.</p>
    490. 

  490. <div class="admonition note">
    491. 

  491. <p class="admonition-title">Note</p>
    492. 

  492. <p>Although the recipe for forward pass needs to be defined within
    493. 

  493. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    494. 

  494. instead of this since the former takes care of running the
    495. 

  495. registered hooks while the latter silently ignores them.</p>
    496. 

  496. </div>
    497. 

  497. </dd></dl>
    498. 

  498. 

  499. <dl class="py attribute">
    500. 

  500. <dt class="sig sig-object py" id="super_gradients.training.losses.FocalLoss.training">
    501. 

  501. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.FocalLoss.training" title="Permalink to this definition"></a></dt>
    502. 

  502. <dd></dd></dl>
    503. 

  503. 

  504. </dd></dl>
    505. 

  505. 

  506. <dl class="py class">
    507. 

  507. <dt class="sig sig-object py" id="super_gradients.training.losses.LabelSmoothingCrossEntropyLoss">
    508. 

  508. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">LabelSmoothingCrossEntropyLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">weight</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_index</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">-</span> <span class="pre">100</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">reduction</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">'mean'</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">smooth_eps</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">smooth_dist</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">from_logits</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/label_smoothing_cross_entropy_loss.html#LabelSmoothingCrossEntropyLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.LabelSmoothingCrossEntropyLoss" title="Permalink to this definition"></a></dt>
    509. 

  509. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss.CrossEntropyLoss</span></code></p>
    510. 

  510. <p>CrossEntropyLoss - with ability to recieve distrbution as targets, and optional label smoothing</p>
    511. 

  511. <dl class="py method">
    512. 

  512. <dt class="sig sig-object py" id="super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.forward">
    513. 

  513. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">input</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">target</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">smooth_dist</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/label_smoothing_cross_entropy_loss.html#LabelSmoothingCrossEntropyLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.forward" title="Permalink to this definition"></a></dt>
    514. 

  514. <dd><p>Defines the computation performed at every call.</p>
    515. 

  515. <p>Should be overridden by all subclasses.</p>
    516. 

  516. <div class="admonition note">
    517. 

  517. <p class="admonition-title">Note</p>
    518. 

  518. <p>Although the recipe for forward pass needs to be defined within
    519. 

  519. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    520. 

  520. instead of this since the former takes care of running the
    521. 

  521. registered hooks while the latter silently ignores them.</p>
    522. 

  522. </div>
    523. 

  523. </dd></dl>
    524. 

  524. 

  525. <dl class="py attribute">
    526. 

  526. <dt class="sig sig-object py" id="super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.ignore_index">
    527. 

  527. <span class="sig-name descname"><span class="pre">ignore_index</span></span><em class="property"><span class="pre">:</span> <span class="pre">int</span></em><a class="headerlink" href="#super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.ignore_index" title="Permalink to this definition"></a></dt>
    528. 

  528. <dd></dd></dl>
    529. 

  529. 

  530. <dl class="py attribute">
    531. 

  531. <dt class="sig sig-object py" id="super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.label_smoothing">
    532. 

  532. <span class="sig-name descname"><span class="pre">label_smoothing</span></span><em class="property"><span class="pre">:</span> <span class="pre">float</span></em><a class="headerlink" href="#super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.label_smoothing" title="Permalink to this definition"></a></dt>
    533. 

  533. <dd></dd></dl>
    534. 

  534. 

  535. <dl class="py attribute">
    536. 

  536. <dt class="sig sig-object py" id="super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.weight">
    537. 

  537. <span class="sig-name descname"><span class="pre">weight</span></span><em class="property"><span class="pre">:</span> <span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">Tensor</span><span class="p"><span class="pre">]</span></span></em><a class="headerlink" href="#super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.weight" title="Permalink to this definition"></a></dt>
    538. 

  538. <dd></dd></dl>
    539. 

  539. 

  540. <dl class="py attribute">
    541. 

  541. <dt class="sig sig-object py" id="super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.reduction">
    542. 

  542. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.reduction" title="Permalink to this definition"></a></dt>
    543. 

  543. <dd></dd></dl>
    544. 

  544. 

  545. <dl class="py attribute">
    546. 

  546. <dt class="sig sig-object py" id="super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.training">
    547. 

  547. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.LabelSmoothingCrossEntropyLoss.training" title="Permalink to this definition"></a></dt>
    548. 

  548. <dd></dd></dl>
    549. 

  549. 

  550. </dd></dl>
    551. 

  551. 

  552. <dl class="py class">
    553. 

  553. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetOHEMLoss">
    554. 

  554. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">ShelfNetOHEMLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">threshold</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.7</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">mining_percent</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.0001</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_lb</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">255</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/shelfnet_ohem_loss.html#ShelfNetOHEMLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ShelfNetOHEMLoss" title="Permalink to this definition"></a></dt>
    555. 

  555. <dd><p>Bases: <a class="reference internal" href="#super_gradients.training.losses.ohem_ce_loss.OhemCELoss" title="super_gradients.training.losses.ohem_ce_loss.OhemCELoss"><code class="xref py py-class docutils literal notranslate"><span class="pre">super_gradients.training.losses.ohem_ce_loss.OhemCELoss</span></code></a></p>
    556. 

  556. <dl class="py method">
    557. 

  557. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetOHEMLoss.forward">
    558. 

  558. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">predictions_list</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">list</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">targets</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/shelfnet_ohem_loss.html#ShelfNetOHEMLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ShelfNetOHEMLoss.forward" title="Permalink to this definition"></a></dt>
    559. 

  559. <dd><p>Defines the computation performed at every call.</p>
    560. 

  560. <p>Should be overridden by all subclasses.</p>
    561. 

  561. <div class="admonition note">
    562. 

  562. <p class="admonition-title">Note</p>
    563. 

  563. <p>Although the recipe for forward pass needs to be defined within
    564. 

  564. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    565. 

  565. instead of this since the former takes care of running the
    566. 

  566. registered hooks while the latter silently ignores them.</p>
    567. 

  567. </div>
    568. 

  568. </dd></dl>
    569. 

  569. 

  570. <dl class="py attribute">
    571. 

  571. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetOHEMLoss.reduction">
    572. 

  572. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.ShelfNetOHEMLoss.reduction" title="Permalink to this definition"></a></dt>
    573. 

  573. <dd></dd></dl>
    574. 

  574. 

  575. <dl class="py attribute">
    576. 

  576. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetOHEMLoss.training">
    577. 

  577. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.ShelfNetOHEMLoss.training" title="Permalink to this definition"></a></dt>
    578. 

  578. <dd></dd></dl>
    579. 

  579. 

  580. </dd></dl>
    581. 

  581. 

  582. <dl class="py class">
    583. 

  583. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetSemanticEncodingLoss">
    584. 

  584. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">ShelfNetSemanticEncodingLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">se_weight</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.2</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">nclass</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">21</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">aux_weight</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">0.4</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">weight</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_index</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">-</span> <span class="pre">1</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/shelfnet_semantic_encoding_loss.html#ShelfNetSemanticEncodingLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ShelfNetSemanticEncodingLoss" title="Permalink to this definition"></a></dt>
    585. 

  585. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss.CrossEntropyLoss</span></code></p>
    586. 

  586. <p>2D Cross Entropy Loss with Auxilary Loss</p>
    587. 

  587. <dl class="py method">
    588. 

  588. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetSemanticEncodingLoss.forward">
    589. 

  589. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">logits</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">labels</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/shelfnet_semantic_encoding_loss.html#ShelfNetSemanticEncodingLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.ShelfNetSemanticEncodingLoss.forward" title="Permalink to this definition"></a></dt>
    590. 

  590. <dd><p>Defines the computation performed at every call.</p>
    591. 

  591. <p>Should be overridden by all subclasses.</p>
    592. 

  592. <div class="admonition note">
    593. 

  593. <p class="admonition-title">Note</p>
    594. 

  594. <p>Although the recipe for forward pass needs to be defined within
    595. 

  595. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    596. 

  596. instead of this since the former takes care of running the
    597. 

  597. registered hooks while the latter silently ignores them.</p>
    598. 

  598. </div>
    599. 

  599. </dd></dl>
    600. 

  600. 

  601. <dl class="py attribute">
    602. 

  602. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetSemanticEncodingLoss.ignore_index">
    603. 

  603. <span class="sig-name descname"><span class="pre">ignore_index</span></span><em class="property"><span class="pre">:</span> <span class="pre">int</span></em><a class="headerlink" href="#super_gradients.training.losses.ShelfNetSemanticEncodingLoss.ignore_index" title="Permalink to this definition"></a></dt>
    604. 

  604. <dd></dd></dl>
    605. 

  605. 

  606. <dl class="py attribute">
    607. 

  607. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetSemanticEncodingLoss.label_smoothing">
    608. 

  608. <span class="sig-name descname"><span class="pre">label_smoothing</span></span><em class="property"><span class="pre">:</span> <span class="pre">float</span></em><a class="headerlink" href="#super_gradients.training.losses.ShelfNetSemanticEncodingLoss.label_smoothing" title="Permalink to this definition"></a></dt>
    609. 

  609. <dd></dd></dl>
    610. 

  610. 

  611. <dl class="py attribute">
    612. 

  612. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetSemanticEncodingLoss.weight">
    613. 

  613. <span class="sig-name descname"><span class="pre">weight</span></span><em class="property"><span class="pre">:</span> <span class="pre">Optional</span><span class="p"><span class="pre">[</span></span><span class="pre">Tensor</span><span class="p"><span class="pre">]</span></span></em><a class="headerlink" href="#super_gradients.training.losses.ShelfNetSemanticEncodingLoss.weight" title="Permalink to this definition"></a></dt>
    614. 

  614. <dd></dd></dl>
    615. 

  615. 

  616. <dl class="py attribute">
    617. 

  617. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetSemanticEncodingLoss.reduction">
    618. 

  618. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.ShelfNetSemanticEncodingLoss.reduction" title="Permalink to this definition"></a></dt>
    619. 

  619. <dd></dd></dl>
    620. 

  620. 

  621. <dl class="py attribute">
    622. 

  622. <dt class="sig sig-object py" id="super_gradients.training.losses.ShelfNetSemanticEncodingLoss.training">
    623. 

  623. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.ShelfNetSemanticEncodingLoss.training" title="Permalink to this definition"></a></dt>
    624. 

  624. <dd></dd></dl>
    625. 

  625. 

  626. </dd></dl>
    627. 

  627. 

  628. <dl class="py class">
    629. 

  629. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss">
    630. 

  630. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">YoloXDetectionLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">strides</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">list</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_classes</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">use_l1</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">bool</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">False</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">center_sampling_radius</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">2.5</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">iou_type</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">'iou'</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/yolox_loss.html#YoloXDetectionLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss" title="Permalink to this definition"></a></dt>
    631. 

  631. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    632. 

  632. <p>Calculate YOLOX loss:
    633. 

  633. L = L_objectivness + L_iou + L_classification + 1[use_l1]*L_l1</p>
    634. 

  634. <dl>
    635. 

  635. <dt>where:</dt><dd><ul class="simple">
    636. 

  636. <li><p>L_iou, L_classification and L_l1 are calculated only between cells and targets that suit them;</p></li>
    637. 

  637. <li><p>L_objectivness is calculated for all cells.</p></li>
    638. 

  638. </ul>
    639. 

  639. <dl class="simple">
    640. 

  640. <dt>L_classification:</dt><dd><p>for cells that have suitable ground truths in their grid locations add BCEs
    641. 

  641. to force a prediction of IoU with a GT in a multi-label way
    642. 

  642. Coef: 1.</p>
    643. 

  643. </dd>
    644. 

  644. <dt>L_iou:</dt><dd><p>for cells that have suitable ground truths in their grid locations
    645. 

  645. add (1 - IoU^2), IoU between a predicted box and each GT box, force maximum IoU
    646. 

  646. Coef: 5.</p>
    647. 

  647. </dd>
    648. 

  648. <dt>L_l1:</dt><dd><p>for cells that have suitable ground truths in their grid locations
    649. 

  649. l1 distance between the logits and GTs in “logits” format (the inverse of “logits to predictions” ops)
    650. 

  650. Coef: 1[use_l1]</p>
    651. 

  651. </dd>
    652. 

  652. <dt>L_objectness:</dt><dd><p>for each cell add BCE with a label of 1 if there is GT assigned to the cell
    653. 

  653. Coef: 1</p>
    654. 

  654. </dd>
    655. 

  655. </dl>
    656. 

  656. </dd>
    657. 

  657. </dl>
    658. 

  658. <dl class="py attribute">
    659. 

  659. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.strides">
    660. 

  660. <span class="sig-name descname"><span class="pre">strides</span></span><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.strides" title="Permalink to this definition"></a></dt>
    661. 

  661. <dd><p>list: List of Yolo levels output grid sizes (i.e [8, 16, 32]).</p>
    662. 

  662. </dd></dl>
    663. 

  663. 

  664. <dl class="py attribute">
    665. 

  665. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.num_classes">
    666. 

  666. <span class="sig-name descname"><span class="pre">num_classes</span></span><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.num_classes" title="Permalink to this definition"></a></dt>
    667. 

  667. <dd><p>int: Number of classes.</p>
    668. 

  668. </dd></dl>
    669. 

  669. 

  670. <dl class="py attribute">
    671. 

  671. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.use_l1">
    672. 

  672. <span class="sig-name descname"><span class="pre">use_l1</span></span><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.use_l1" title="Permalink to this definition"></a></dt>
    673. 

  673. <dd><p>bool: Controls the L_l1 Coef as discussed above (default=False).</p>
    674. 

  674. </dd></dl>
    675. 

  675. 

  676. <dl class="py attribute">
    677. 

  677. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.center_sampling_radius">
    678. 

  678. <span class="sig-name descname"><span class="pre">center_sampling_radius</span></span><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.center_sampling_radius" title="Permalink to this definition"></a></dt>
    679. 

  679. <dd><p>float: Sampling radius used for center sampling when creating the fg mask (default=2.5).</p>
    680. 

  680. </dd></dl>
    681. 

  681. 

  682. <dl class="py attribute">
    683. 

  683. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.iou_type">
    684. 

  684. <span class="sig-name descname"><span class="pre">iou_type</span></span><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.iou_type" title="Permalink to this definition"></a></dt>
    685. 

  685. <dd><p>str: Iou loss type, one of [“iou”,”giou”] (deafult=”iou”).</p>
    686. 

  686. </dd></dl>
    687. 

  687. 

  688. <dl class="py method">
    689. 

  689. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.forward">
    690. 

  690. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">model_output</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">list</span><span class="p"><span class="pre">,</span> </span><span class="pre">Tuple</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.Tensor</span><span class="p"><span class="pre">,</span> </span><span class="pre">List</span><span class="p"><span class="pre">]</span></span><span class="p"><span class="pre">]</span></span></span></em>, <em class="sig-param"><span class="n"><span class="pre">targets</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.Tensor</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/yolox_loss.html#YoloXDetectionLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.forward" title="Permalink to this definition"></a></dt>
    691. 

  691. <dd><dl class="field-list simple">
    692. 

  692. <dt class="field-odd">Parameters</dt>
    693. 

  693. <dd class="field-odd"><ul class="simple">
    694. 

  694. <li><p><strong>model_output</strong> – <p>Union[list, Tuple[torch.Tensor, List]]:
    695. 

  695. When list-</p>
    696. 

  696. <blockquote>
    697. 

  697. <div><p>output from all Yolo levels, each of shape [Batch x 1 x GridSizeY x GridSizeX x (4 + 1 + Num_classes)]</p>
    698. 

  698. </div></blockquote>
    699. 

  699. <p>And when tuple- the second item is the described list (first item is discarded)</p>
    700. 

  700. </p></li>
    701. 

  701. <li><p><strong>targets</strong> – torch.Tensor: Num_targets x (4 + 2)], values on dim 1 are: image id in a batch, class, box x y w h</p></li>
    702. 

  702. </ul>
    703. 

  703. </dd>
    704. 

  704. <dt class="field-even">Returns</dt>
    705. 

  705. <dd class="field-even"><p>loss, all losses separately in a detached tensor</p>
    706. 

  706. </dd>
    707. 

  707. </dl>
    708. 

  708. </dd></dl>
    709. 

  709. 

  710. <dl class="py method">
    711. 

  711. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.prepare_predictions">
    712. 

  712. <span class="sig-name descname"><span class="pre">prepare_predictions</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">predictions</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">List</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.Tensor</span><span class="p"><span class="pre">]</span></span></span></em><span class="sig-paren">)</span> &#x2192; <span class="pre">Tuple</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.Tensor</span><span class="p"><span class="pre">,</span> </span><span class="pre">torch.Tensor</span><span class="p"><span class="pre">,</span> </span><span class="pre">torch.Tensor</span><span class="p"><span class="pre">,</span> </span><span class="pre">torch.Tensor</span><span class="p"><span class="pre">,</span> </span><span class="pre">torch.Tensor</span><span class="p"><span class="pre">]</span></span><a class="reference internal" href="_modules/super_gradients/training/losses/yolox_loss.html#YoloXDetectionLoss.prepare_predictions"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.prepare_predictions" title="Permalink to this definition"></a></dt>
    713. 

  713. <dd><p>Convert raw outputs of the network into a format that merges outputs from all levels
    714. 

  714. :param predictions:     output from all Yolo levels, each of shape</p>
    715. 

  715. <blockquote>
    716. 

  716. <div><p>[Batch x 1 x GridSizeY x GridSizeX x (4 + 1 + Num_classes)]</p>
    717. 

  717. </div></blockquote>
    718. 

  718. <dl class="field-list simple">
    719. 

  719. <dt class="field-odd">Returns</dt>
    720. 

  720. <dd class="field-odd"><p><p>5 tensors representing predictions:
    721. 

  721. * x_shifts: shape [1 x * num_cells x 1],</p>
    722. 

  722. <blockquote>
    723. 

  723. <div><p>where num_cells = grid1X * grid1Y + grid2X * grid2Y + grid3X * grid3Y,
    724. 

  724. x coordinate on the grid cell the prediction is coming from</p>
    725. 

  725. </div></blockquote>
    726. 

  726. <ul class="simple">
    727. 

  727. <li><p>y_shifts: shape [1 x  num_cells x 1],
    728. 

  728. y coordinate on the grid cell the prediction is coming from</p></li>
    729. 

  729. <li><p>expanded_strides: shape [1 x num_cells x 1],
    730. 

  730. stride of the output grid the prediction is coming from</p></li>
    731. 

  731. <li><p>transformed_outputs: shape [batch_size x num_cells x (num_classes + 5)],
    732. 

  732. predictions with boxes in real coordinates and logprobabilities</p></li>
    733. 

  733. <li><p>raw_outputs: shape [batch_size x num_cells x (num_classes + 5)],
    734. 

  734. raw predictions with boxes and confidences as logits</p></li>
    735. 

  735. </ul>
    736. 

  736. </p>
    737. 

  737. </dd>
    738. 

  738. </dl>
    739. 

  739. </dd></dl>
    740. 

  740. 

  741. <dl class="py method">
    742. 

  742. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.get_l1_target">
    743. 

  743. <span class="sig-name descname"><span class="pre">get_l1_target</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">l1_target</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">gt</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">stride</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">x_shifts</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">y_shifts</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">eps</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">1e-08</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/yolox_loss.html#YoloXDetectionLoss.get_l1_target"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.get_l1_target" title="Permalink to this definition"></a></dt>
    744. 

  744. <dd><dl class="field-list simple">
    745. 

  745. <dt class="field-odd">Parameters</dt>
    746. 

  746. <dd class="field-odd"><ul class="simple">
    747. 

  747. <li><p><strong>l1_target</strong> – tensor of zeros of shape [Num_cell_gt_pairs x 4]</p></li>
    748. 

  748. <li><p><strong>gt</strong> – targets in coordinates [Num_cell_gt_pairs x (4 + 1 + num_classes)]</p></li>
    749. 

  749. </ul>
    750. 

  750. </dd>
    751. 

  751. <dt class="field-even">Returns</dt>
    752. 

  752. <dd class="field-even"><p>targets in the format corresponding to logits</p>
    753. 

  753. </dd>
    754. 

  754. </dl>
    755. 

  755. </dd></dl>
    756. 

  756. 

  757. <dl class="py method">
    758. 

  758. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.get_assignments">
    759. 

  759. <span class="sig-name descname"><span class="pre">get_assignments</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">image_idx</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_gt</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">total_num_anchors</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">gt_bboxes_per_image</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">gt_classes</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">bboxes_preds_per_image</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">expanded_strides</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">x_shifts</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">y_shifts</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">cls_preds</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">obj_preds</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">mode</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">'gpu'</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ious_loss_cost_coeff</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">3.0</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">outside_boxes_and_center_cost_coeff</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">100000.0</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/yolox_loss.html#YoloXDetectionLoss.get_assignments"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.get_assignments" title="Permalink to this definition"></a></dt>
    760. 

  760. <dd><dl class="simple">
    761. 

  761. <dt>Match cells to ground truth:</dt><dd><ul class="simple">
    762. 

  762. <li><p>at most 1 GT per cell</p></li>
    763. 

  763. <li><p>dynamic number of cells per GT</p></li>
    764. 

  764. </ul>
    765. 

  765. </dd>
    766. 

  766. </dl>
    767. 

  767. <dl class="field-list simple">
    768. 

  768. <dt class="field-odd">Parameters</dt>
    769. 

  769. <dd class="field-odd"><ul class="simple">
    770. 

  770. <li><p><strong>outside_boxes_and_center_cost_coeff</strong> – float: Cost coefficiant of cells the radius and bbox of gts in dynamic
    771. 

  771. matching (default=100000).</p></li>
    772. 

  772. <li><p><strong>ious_loss_cost_coeff</strong> – float: Cost coefficiant for iou loss in dynamic matching (default=3).</p></li>
    773. 

  773. <li><p><strong>image_idx</strong> – int: Image index in batch.</p></li>
    774. 

  774. <li><p><strong>num_gt</strong> – int: Number of ground trunth targets in the image.</p></li>
    775. 

  775. <li><p><strong>total_num_anchors</strong> – int: Total number of possible bboxes = sum of all grid cells.</p></li>
    776. 

  776. <li><p><strong>gt_bboxes_per_image</strong> – torch.Tensor: Tensor of gt bboxes for  the image, shape: (num_gt, 4).</p></li>
    777. 

  777. <li><p><strong>gt_classes</strong> – torch.Tesnor: Tensor of the classes in the image, shape: (num_preds,4).</p></li>
    778. 

  778. <li><p><strong>bboxes_preds_per_image</strong> – Tensor of the classes in the image, shape: (num_preds).</p></li>
    779. 

  779. <li><p><strong>expanded_strides</strong> – torch.Tensor: Stride of the output grid the prediction is coming from,
    780. 

  780. shape (1 x num_cells x 1).</p></li>
    781. 

  781. <li><p><strong>x_shifts</strong> – torch.Tensor: X’s in cell coordinates, shape (1,num_cells,1).</p></li>
    782. 

  782. <li><p><strong>y_shifts</strong> – torch.Tensor: Y’s in cell coordinates, shape (1,num_cells,1).</p></li>
    783. 

  783. <li><p><strong>cls_preds</strong> – torch.Tensor: Class predictions in all cells, shape (batch_size, num_cells).</p></li>
    784. 

  784. <li><p><strong>obj_preds</strong> – torch.Tensor: Objectness predictions in all cells, shape (batch_size, num_cells).</p></li>
    785. 

  785. <li><p><strong>mode</strong> – str: One of [“gpu”,”cpu”], Controls the device the assignment operation should be taken place on (deafult=”gpu”)</p></li>
    786. 

  786. </ul>
    787. 

  787. </dd>
    788. 

  788. </dl>
    789. 

  789. </dd></dl>
    790. 

  790. 

  791. <dl class="py method">
    792. 

  792. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.get_in_boxes_info">
    793. 

  793. <span class="sig-name descname"><span class="pre">get_in_boxes_info</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">gt_bboxes_per_image</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">expanded_strides</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">x_shifts</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">y_shifts</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">total_num_anchors</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_gt</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/yolox_loss.html#YoloXDetectionLoss.get_in_boxes_info"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.get_in_boxes_info" title="Permalink to this definition"></a></dt>
    794. 

  794. <dd><dl>
    795. 

  795. <dt>Create a mask for all cells, mask in only foreground: cells that have a center located:</dt><dd><ul class="simple">
    796. 

  796. <li><p>withing a GT box;</p></li>
    797. 

  797. </ul>
    798. 

  798. <p>OR
    799. 

  799. * within a fixed radius around a GT box (center sampling);</p>
    800. 

  800. </dd>
    801. 

  801. </dl>
    802. 

  802. <dl class="field-list simple">
    803. 

  803. <dt class="field-odd">Parameters</dt>
    804. 

  804. <dd class="field-odd"><ul class="simple">
    805. 

  805. <li><p><strong>num_gt</strong> – int: Number of ground trunth targets in the image.</p></li>
    806. 

  806. <li><p><strong>total_num_anchors</strong> – int: Sum of all grid cells.</p></li>
    807. 

  807. <li><p><strong>gt_bboxes_per_image</strong> – torch.Tensor: Tensor of gt bboxes for  the image, shape: (num_gt, 4).</p></li>
    808. 

  808. <li><p><strong>expanded_strides</strong> – torch.Tensor: Stride of the output grid the prediction is coming from,
    809. 

  809. shape (1 x num_cells x 1).</p></li>
    810. 

  810. <li><p><strong>x_shifts</strong> – torch.Tensor: X’s in cell coordinates, shape (1,num_cells,1).</p></li>
    811. 

  811. <li><p><strong>y_shifts</strong> – torch.Tensor: Y’s in cell coordinates, shape (1,num_cells,1).</p></li>
    812. 

  812. </ul>
    813. 

  813. </dd>
    814. 

  814. </dl>
    815. 

  815. <dl class="simple">
    816. 

  816. <dt>:return is_in_boxes_anchor, is_in_boxes_and_center</dt><dd><dl class="simple">
    817. 

  817. <dt>where:</dt><dd><ul class="simple">
    818. 

  818. <li><dl class="simple">
    819. 

  819. <dt>is_in_boxes_anchor masks the cells that their cell center is  inside a gt bbox and within</dt><dd><p>self.center_sampling_radius cells away, without reduction (i.e shape=(num_gts, num_fgs))</p>
    820. 

  820. </dd>
    821. 

  821. </dl>
    822. 

  822. </li>
    823. 

  823. <li><dl class="simple">
    824. 

  824. <dt>is_in_boxes_and_center masks the cells that their center is either inside a gt bbox or within</dt><dd><p>self.center_sampling_radius cells away, shape (num_fgs)</p>
    825. 

  825. </dd>
    826. 

  826. </dl>
    827. 

  827. </li>
    828. 

  828. </ul>
    829. 

  829. </dd>
    830. 

  830. </dl>
    831. 

  831. </dd>
    832. 

  832. </dl>
    833. 

  833. </dd></dl>
    834. 

  834. 

  835. <dl class="py method">
    836. 

  836. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.dynamic_k_matching">
    837. 

  837. <span class="sig-name descname"><span class="pre">dynamic_k_matching</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">cost</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">pair_wise_ious</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">gt_classes</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_gt</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">fg_mask</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/yolox_loss.html#YoloXDetectionLoss.dynamic_k_matching"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.dynamic_k_matching" title="Permalink to this definition"></a></dt>
    838. 

  838. <dd><dl class="field-list simple">
    839. 

  839. <dt class="field-odd">Parameters</dt>
    840. 

  840. <dd class="field-odd"><ul class="simple">
    841. 

  841. <li><p><strong>cost</strong> – pairwise cost, [num_FGs x num_GTs]</p></li>
    842. 

  842. <li><p><strong>pair_wise_ious</strong> – pairwise IoUs, [num_FGs x num_GTs]</p></li>
    843. 

  843. <li><p><strong>gt_classes</strong> – class of each GT</p></li>
    844. 

  844. <li><p><strong>num_gt</strong> – number of GTs</p></li>
    845. 

  845. </ul>
    846. 

  846. </dd>
    847. 

  847. </dl>
    848. 

  848. <dl class="simple">
    849. 

  849. <dt>:return num_fg, (number of foregrounds)</dt><dd><p>gt_matched_classes, (the classes that have been matched with fgs)
    850. 

  850. pred_ious_this_matching
    851. 

  851. matched_gt_inds</p>
    852. 

  852. </dd>
    853. 

  853. </dl>
    854. 

  854. </dd></dl>
    855. 

  855. 

  856. <dl class="py attribute">
    857. 

  857. <dt class="sig sig-object py" id="super_gradients.training.losses.YoloXDetectionLoss.reduction">
    858. 

  858. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.YoloXDetectionLoss.reduction" title="Permalink to this definition"></a></dt>
    859. 

  859. <dd></dd></dl>
    860. 

  860. 

  861. </dd></dl>
    862. 

  862. 

  863. <dl class="py class">
    864. 

  864. <dt class="sig sig-object py" id="super_gradients.training.losses.RSquaredLoss">
    865. 

  865. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">RSquaredLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">size_average</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">reduce</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">None</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">reduction</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">str</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">'mean'</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/r_squared_loss.html#RSquaredLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.RSquaredLoss" title="Permalink to this definition"></a></dt>
    866. 

  866. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    867. 

  867. <dl class="py method">
    868. 

  868. <dt class="sig sig-object py" id="super_gradients.training.losses.RSquaredLoss.forward">
    869. 

  869. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">output</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">target</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/r_squared_loss.html#RSquaredLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.RSquaredLoss.forward" title="Permalink to this definition"></a></dt>
    870. 

  870. <dd><p>Computes the R-squared for the output and target values
    871. 

  871. :param output: Tensor / Numpy / List</p>
    872. 

  872. <blockquote>
    873. 

  873. <div><p>The prediction</p>
    874. 

  874. </div></blockquote>
    875. 

  875. <dl class="field-list simple">
    876. 

  876. <dt class="field-odd">Parameters</dt>
    877. 

  877. <dd class="field-odd"><p><strong>target</strong> – Tensor / Numpy / List
    878. 

  878. The corresponding lables</p>
    879. 

  879. </dd>
    880. 

  880. </dl>
    881. 

  881. </dd></dl>
    882. 

  882. 

  883. <dl class="py attribute">
    884. 

  884. <dt class="sig sig-object py" id="super_gradients.training.losses.RSquaredLoss.reduction">
    885. 

  885. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.RSquaredLoss.reduction" title="Permalink to this definition"></a></dt>
    886. 

  886. <dd></dd></dl>
    887. 

  887. 

  888. <dl class="py attribute">
    889. 

  889. <dt class="sig sig-object py" id="super_gradients.training.losses.RSquaredLoss.training">
    890. 

  890. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.RSquaredLoss.training" title="Permalink to this definition"></a></dt>
    891. 

  891. <dd></dd></dl>
    892. 

  892. 

  893. </dd></dl>
    894. 

  894. 

  895. <dl class="py class">
    896. 

  896. <dt class="sig sig-object py" id="super_gradients.training.losses.SSDLoss">
    897. 

  897. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">SSDLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">dboxes</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><a class="reference internal" href="super_gradients.training.utils.html#super_gradients.training.utils.ssd_utils.DefaultBoxes" title="super_gradients.training.utils.ssd_utils.DefaultBoxes"><span class="pre">super_gradients.training.utils.ssd_utils.DefaultBoxes</span></a></span></em>, <em class="sig-param"><span class="n"><span class="pre">alpha</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">1.0</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">iou_thresh</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.5</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">neg_pos_ratio</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">3.0</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ssd_loss.html#SSDLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.SSDLoss" title="Permalink to this definition"></a></dt>
    898. 

  898. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    899. 

  899. <blockquote>
    900. 

  900. <div><p>Implements the loss as the sum of the followings:
    901. 

  901. 1. Confidence Loss: All labels, with hard negative mining
    902. 

  902. 2. Localization Loss: Only on positive labels</p>
    903. 

  903. </div></blockquote>
    904. 

  904. <dl class="simple">
    905. 

  905. <dt>L = (2 - alpha) * L_l1 + alpha * L_cls, where</dt><dd><ul class="simple">
    906. 

  906. <li><p>L_cls is HardMiningCrossEntropyLoss</p></li>
    907. 

  907. <li><p>L_l1 = [SmoothL1Loss for all positives]</p></li>
    908. 

  908. </ul>
    909. 

  909. </dd>
    910. 

  910. </dl>
    911. 

  911. <dl class="py method">
    912. 

  912. <dt class="sig sig-object py" id="super_gradients.training.losses.SSDLoss.match_dboxes">
    913. 

  913. <span class="sig-name descname"><span class="pre">match_dboxes</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">targets</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ssd_loss.html#SSDLoss.match_dboxes"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.SSDLoss.match_dboxes" title="Permalink to this definition"></a></dt>
    914. 

  914. <dd><p>creates tensors with target boxes and labels for each dboxes, so with the same len as dboxes.</p>
    915. 

  915. <ul class="simple">
    916. 

  916. <li><p>Each GT is assigned with a grid cell with the highest IoU, this creates a pair for each GT and some cells;</p></li>
    917. 

  917. <li><p>The rest of grid cells are assigned to a GT with the highest IoU, assuming it’s &gt; self.iou_thresh;
    918. 

  918. If this condition is not met the grid cell is marked as background</p></li>
    919. 

  919. </ul>
    920. 

  920. <p>GT-wise: one to many
    921. 

  921. Grid-cell-wise: one to one</p>
    922. 

  922. <dl class="field-list simple">
    923. 

  923. <dt class="field-odd">Parameters</dt>
    924. 

  924. <dd class="field-odd"><p><strong>targets</strong> – a tensor containing the boxes for a single image;
    925. 

  925. shape [num_boxes, 6] (image_id, label, x, y, w, h)</p>
    926. 

  926. </dd>
    927. 

  927. <dt class="field-even">Returns</dt>
    928. 

  928. <dd class="field-even"><p>two tensors
    929. 

  929. boxes - shape of dboxes [4, num_dboxes] (x,y,w,h)
    930. 

  930. labels - sahpe [num_dboxes]</p>
    931. 

  931. </dd>
    932. 

  932. </dl>
    933. 

  933. </dd></dl>
    934. 

  934. 

  935. <dl class="py method">
    936. 

  936. <dt class="sig sig-object py" id="super_gradients.training.losses.SSDLoss.forward">
    937. 

  937. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">predictions</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Tuple</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">targets</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/ssd_loss.html#SSDLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.SSDLoss.forward" title="Permalink to this definition"></a></dt>
    938. 

  938. <dd><dl class="simple">
    939. 

  939. <dt>Compute the loss</dt><dd><p>:param predictions - predictions tensor coming from the network,
    940. 

  940. tuple with shapes ([Batch Size, 4, num_dboxes], [Batch Size, num_classes + 1, num_dboxes])
    941. 

  941. were predictions have logprobs for background and other classes
    942. 

  942. :param targets - targets for the batch. [num targets, 6] (index in batch, label, x,y,w,h)</p>
    943. 

  943. </dd>
    944. 

  944. </dl>
    945. 

  945. </dd></dl>
    946. 

  946. 

  947. <dl class="py attribute">
    948. 

  948. <dt class="sig sig-object py" id="super_gradients.training.losses.SSDLoss.reduction">
    949. 

  949. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.SSDLoss.reduction" title="Permalink to this definition"></a></dt>
    950. 

  950. <dd></dd></dl>
    951. 

  951. 

  952. <dl class="py attribute">
    953. 

  953. <dt class="sig sig-object py" id="super_gradients.training.losses.SSDLoss.training">
    954. 

  954. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.SSDLoss.training" title="Permalink to this definition"></a></dt>
    955. 

  955. <dd></dd></dl>
    956. 

  956. 

  957. </dd></dl>
    958. 

  958. 

  959. <dl class="py class">
    960. 

  960. <dt class="sig sig-object py" id="super_gradients.training.losses.BCEDiceLoss">
    961. 

  961. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">BCEDiceLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">loss_weights</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">[0.5,</span> <span class="pre">0.5]</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">logits</span></span><span class="o"><span class="pre">=</span></span><span class="default_value"><span class="pre">True</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/bce_dice_loss.html#BCEDiceLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.BCEDiceLoss" title="Permalink to this definition"></a></dt>
    962. 

  962. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.module.Module</span></code></p>
    963. 

  963. <p>Binary Cross Entropy + Dice Loss</p>
    964. 

  964. <p>Weighted average of BCE and Dice loss</p>
    965. 

  965. <dl class="py attribute">
    966. 

  966. <dt class="sig sig-object py" id="super_gradients.training.losses.BCEDiceLoss.loss_weights">
    967. 

  967. <span class="sig-name descname"><span class="pre">loss_weights</span></span><a class="headerlink" href="#super_gradients.training.losses.BCEDiceLoss.loss_weights" title="Permalink to this definition"></a></dt>
    968. 

  968. <dd><p>list of size 2 s.t loss_weights[0], loss_weights[1] are the weights for BCE, Dice</p>
    969. 

  969. </dd></dl>
    970. 

  970. 

  971. <dl class="py attribute">
    972. 

  972. <dt class="sig sig-object py">
    973. 

  973. <span class="sig-name descname"><span class="pre">respectively.</span></span></dt>
    974. 

  974. <dd></dd></dl>
    975. 

  975. 

  976. <dl class="py method">
    977. 

  977. <dt class="sig sig-object py" id="super_gradients.training.losses.BCEDiceLoss.forward">
    978. 

  978. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">input</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.Tensor</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">target</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.Tensor</span></span></em><span class="sig-paren">)</span> &#x2192; <span class="pre">torch.Tensor</span><a class="reference internal" href="_modules/super_gradients/training/losses/bce_dice_loss.html#BCEDiceLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.BCEDiceLoss.forward" title="Permalink to this definition"></a></dt>
    979. 

  979. <dd><p>&#64;param input: Network’s raw output shaped (N,1,H,W)
    980. 

  980. &#64;param target: Ground truth shaped (N,H,W)</p>
    981. 

  981. </dd></dl>
    982. 

  982. 

  983. <dl class="py attribute">
    984. 

  984. <dt class="sig sig-object py" id="super_gradients.training.losses.BCEDiceLoss.training">
    985. 

  985. <span class="sig-name descname"><span class="pre">training</span></span><em class="property"><span class="pre">:</span> <span class="pre">bool</span></em><a class="headerlink" href="#super_gradients.training.losses.BCEDiceLoss.training" title="Permalink to this definition"></a></dt>
    986. 

  986. <dd></dd></dl>
    987. 

  987. 

  988. </dd></dl>
    989. 

  989. 

  990. <dl class="py class">
    991. 

  991. <dt class="sig sig-object py" id="super_gradients.training.losses.KDLogitsLoss">
    992. 

  992. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">KDLogitsLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">task_loss_fn</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.nn.modules.loss._Loss</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">distillation_loss_fn</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.nn.modules.loss._Loss</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">KDklDivLoss()</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">distillation_loss_coeff</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">float</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">0.5</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/kd_losses.html#KDLogitsLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.KDLogitsLoss" title="Permalink to this definition"></a></dt>
    993. 

  993. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    994. 

  994. <p>Knowledge distillation loss, wraps the task loss and distillation loss</p>
    995. 

  995. <dl class="py method">
    996. 

  996. <dt class="sig sig-object py" id="super_gradients.training.losses.KDLogitsLoss.forward">
    997. 

  997. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">kd_module_output</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">target</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/kd_losses.html#KDLogitsLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.KDLogitsLoss.forward" title="Permalink to this definition"></a></dt>
    998. 

  998. <dd><p>Defines the computation performed at every call.</p>
    999. 

  999. <p>Should be overridden by all subclasses.</p>
    1000. 

  1000. <div class="admonition note">
    1001. 

  1001. <p class="admonition-title">Note</p>
    1002. 

  1002. <p>Although the recipe for forward pass needs to be defined within
    1003. 

  1003. this function, one should call the <code class="xref py py-class docutils literal notranslate"><span class="pre">Module</span></code> instance afterwards
    1004. 

  1004. instead of this since the former takes care of running the
    1005. 

  1005. registered hooks while the latter silently ignores them.</p>
    1006. 

  1006. </div>
    1007. 

  1007. </dd></dl>
    1008. 

  1008. 

  1009. <dl class="py attribute">
    1010. 

  1010. <dt class="sig sig-object py" id="super_gradients.training.losses.KDLogitsLoss.reduction">
    1011. 

  1011. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.KDLogitsLoss.reduction" title="Permalink to this definition"></a></dt>
    1012. 

  1012. <dd></dd></dl>
    1013. 

  1013. 

  1014. </dd></dl>
    1015. 

  1015. 

  1016. <dl class="py class">
    1017. 

  1017. <dt class="sig sig-object py" id="super_gradients.training.losses.DiceCEEdgeLoss">
    1018. 

  1018. <em class="property"><span class="pre">class</span> </em><span class="sig-prename descclassname"><span class="pre">super_gradients.training.losses.</span></span><span class="sig-name descname"><span class="pre">DiceCEEdgeLoss</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">num_classes</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_aux_heads</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">2</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">num_detail_heads</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">1</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">weights</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">tuple</span><span class="p"><span class="pre">,</span> </span><span class="pre">list</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">(1,</span> <span class="pre">1,</span> <span class="pre">1,</span> <span class="pre">1)</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">dice_ce_weights</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">tuple</span><span class="p"><span class="pre">,</span> </span><span class="pre">list</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">(1,</span> <span class="pre">1)</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ignore_index</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">-</span> <span class="pre">100</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">edge_kernel</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">int</span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">3</span></span></em>, <em class="sig-param"><span class="n"><span class="pre">ce_edge_weights</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Union</span><span class="p"><span class="pre">[</span></span><span class="pre">tuple</span><span class="p"><span class="pre">,</span> </span><span class="pre">list</span><span class="p"><span class="pre">]</span></span></span> <span class="o"><span class="pre">=</span></span> <span class="default_value"><span class="pre">(0.5,</span> <span class="pre">0.5)</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/dice_ce_edge_loss.html#DiceCEEdgeLoss"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.DiceCEEdgeLoss" title="Permalink to this definition"></a></dt>
    1019. 

  1019. <dd><p>Bases: <code class="xref py py-class docutils literal notranslate"><span class="pre">torch.nn.modules.loss._Loss</span></code></p>
    1020. 

  1020. <dl class="py method">
    1021. 

  1021. <dt class="sig sig-object py" id="super_gradients.training.losses.DiceCEEdgeLoss.forward">
    1022. 

  1022. <span class="sig-name descname"><span class="pre">forward</span></span><span class="sig-paren">(</span><em class="sig-param"><span class="n"><span class="pre">preds</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">Tuple</span><span class="p"><span class="pre">[</span></span><span class="pre">torch.Tensor</span><span class="p"><span class="pre">]</span></span></span></em>, <em class="sig-param"><span class="n"><span class="pre">target</span></span><span class="p"><span class="pre">:</span></span> <span class="n"><span class="pre">torch.Tensor</span></span></em><span class="sig-paren">)</span><a class="reference internal" href="_modules/super_gradients/training/losses/dice_ce_edge_loss.html#DiceCEEdgeLoss.forward"><span class="viewcode-link"><span class="pre">[source]</span></span></a><a class="headerlink" href="#super_gradients.training.losses.DiceCEEdgeLoss.forward" title="Permalink to this definition"></a></dt>
    1023. 

  1023. <dd><dl class="field-list simple">
    1024. 

  1024. <dt class="field-odd">Parameters</dt>
    1025. 

  1025. <dd class="field-odd"><p><strong>preds</strong> – Model output predictions, must be in the followed format:
    1026. 

  1026. [Main-feats, Aux-feats[0], …, Aux-feats[num_auxs-1], Detail-feats[0], …, Detail-feats[num_details-1]</p>
    1027. 

  1027. </dd>
    1028. 

  1028. </dl>
    1029. 

  1029. </dd></dl>
    1030. 

  1030. 

  1031. <dl class="py attribute">
    1032. 

  1032. <dt class="sig sig-object py" id="super_gradients.training.losses.DiceCEEdgeLoss.reduction">
    1033. 

  1033. <span class="sig-name descname"><span class="pre">reduction</span></span><em class="property"><span class="pre">:</span> <span class="pre">str</span></em><a class="headerlink" href="#super_gradients.training.losses.DiceCEEdgeLoss.reduction" title="Permalink to this definition"></a></dt>
    1034. 

  1034. <dd></dd></dl>
    1035. 

  1035. 

  1036. </dd></dl>
    1037. 

  1037. 

  1038. </section>
    1039. 

  1039. </section>
    1040. 

  1040. 

  1041. 

  1042.            </div>
    1043. 

  1043.           </div>
    1044. 

  1044.           <footer>
    1045. 

  1045. 

  1046.   <hr/>
    1047. 

  1047. 

  1048.   <div role="contentinfo">
    1049. 

  1049.     <p>&#169; Copyright 2021, SuperGradients team.</p>
    1050. 

  1050.   </div>
    1051. 

  1051. 

  1052.   Built with <a href="https://www.sphinx-doc.org/">Sphinx</a> using a
    1053. 

  1053.     <a href="https://github.com/readthedocs/sphinx_rtd_theme">theme</a>
    1054. 

  1054.     provided by <a href="https://readthedocs.org">Read the Docs</a>.
    1055. 

  1055.    
    1056. 

  1056. 

  1057. </footer>
    1058. 

  1058.         </div>
    1059. 

  1059.       </div>
    1060. 

  1060.     </section>
    1061. 

  1061.   </div>
    1062. 

  1062.   <script>
    1063. 

  1063.       jQuery(function () {
    1064. 

  1064.           SphinxRtdTheme.Navigation.enable(true);
    1065. 

  1065.       });
    1066. 

  1066.   </script> 
    1067. 

  1067. 

  1068. </body>
    1069. 

  1069. </html>
    1070.
Prev
Next
Tip!

Press p or to see the previous file or, n or to see the next file

Comments

Write Preview
Loading...
Integrate AWS S3
Use S3 remote
Select bucket
Access key
Finish

Use AWS S3 as storage!

Browsing data directories saved to S3 is possible with DAGsHub. Let's configure your repository to easily display your data in the context of any commit!

Specify your S3 bucket

Access key (If needed)

Congratulations!

super-gradients is now integrated with AWS S3!

Delete Storage Key

Are you sure you want to delete this access key?

No
Yes
Integrate Google Cloud Storage
Use Google Storage
Select bucket
Upload key
Finish

Use Google Cloud Storage!

Browsing data directories saved to Google Cloud Storage is possible with DAGsHub. Let's configure your repository to easily display your data in the context of any commit!

Specify your Google Storage bucket

Service Account Key

Congratulations!

super-gradients is now integrated with Google Cloud Storage!

Delete Storage Key

Are you sure you want to delete this access key?

No
Yes
Integrate Azure Cloud Storage
Use Azure Storage
Select bucket
Set key
Finish

Use Azure Cloud Storage!

Browsing data directories saved to Azure Cloud Storage is possible with DAGsHub. Let's configure your repository to easily display your data in the context of any commit!

Specify your Azure Storage bucket

Access key (If needed)

Congratulations!

super-gradients is now integrated with Azure Cloud Storage!

Delete Storage Key

Are you sure you want to delete this access key?

No
Yes
Integrate S3 compatible storage
Use S3 like remote
Select bucket
Access key
Finish

Use any S3 compatible storage!

Browsing data directories saved to S3 compatible storage is possible with DAGsHub. Let's configure your repository to easily display your data in the context of any commit!

Specify your S3 bucket

Bucket name cannot be the same as the repository name. Please change one of them.

Access key (If needed)

Congratulations!

super-gradients is now integrated with your S3 compatible storage!

Delete Storage Key

Are you sure you want to delete this access key?

No
Yes