-
Notifications
You must be signed in to change notification settings - Fork 1
/
onnx_import.html
602 lines (473 loc) · 25.9 KB
/
onnx_import.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
<!DOCTYPE html>
<html class="writer-html5" lang="en" >
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>Import ONNX models — N2D2 documentation</title>
<link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
<link rel="stylesheet" href="_static/pygments.css" type="text/css" />
<!--[if lt IE 9]>
<script src="_static/js/html5shiv.min.js"></script>
<![endif]-->
<script type="text/javascript" id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
<script src="_static/jquery.js"></script>
<script src="_static/underscore.js"></script>
<script src="_static/doctools.js"></script>
<script src="_static/language_data.js"></script>
<script async="async" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.7/latest.js?config=TeX-AMS-MML_HTMLorMML"></script>
<script type="text/javascript" src="_static/js/theme.js"></script>
<link rel="author" title="About these documents" href="about.html" />
<link rel="index" title="Index" href="genindex.html" />
<link rel="search" title="Search" href="search.html" />
<link rel="next" title="Train from ONNX models" href="onnx_transfer.html" />
<link rel="prev" title="Obtain ONNX models" href="onnx_convert.html" />
</head>
<body class="wy-body-for-nav">
<div class="wy-grid-for-nav">
<nav data-toggle="wy-nav-shift" class="wy-nav-side">
<div class="wy-side-scroll">
<div class="wy-side-nav-search" >
<a href="index.html" class="icon icon-home" alt="Documentation Home"> N2D2
</a>
<div role="search">
<form id="rtd-search-form" class="wy-form" action="search.html" method="get">
<input type="text" name="q" placeholder="Search docs" />
<input type="hidden" name="check_keywords" value="yes" />
<input type="hidden" name="area" value="default" />
</form>
</div>
</div>
<div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
<p class="caption"><span class="caption-text">Introduction:</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="intro.html">Presentation</a></li>
<li class="toctree-l1"><a class="reference internal" href="about.html">About N2D2-IP</a></li>
<li class="toctree-l1"><a class="reference internal" href="simus.html">Performing simulations</a></li>
<li class="toctree-l1"><a class="reference internal" href="perfs_tools.html">Performance evaluation tools</a></li>
<li class="toctree-l1"><a class="reference internal" href="tuto.html">Tutorials</a></li>
</ul>
<p class="caption"><span class="caption-text">ONNX Import:</span></p>
<ul class="current">
<li class="toctree-l1"><a class="reference internal" href="onnx_convert.html">Obtain ONNX models</a></li>
<li class="toctree-l1 current"><a class="current reference internal" href="#">Import ONNX models</a><ul>
<li class="toctree-l2"><a class="reference internal" href="#preliminary-steps">Preliminary steps</a></li>
<li class="toctree-l2"><a class="reference internal" href="#with-an-ini-file">With an INI file</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#onnx-ini-section-type">ONNX INI section type</a></li>
<li class="toctree-l3"><a class="reference internal" href="#transpose-option-usage"><code class="docutils literal notranslate"><span class="pre">Transpose</span></code> option usage</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#with-the-python-api">With the Python API</a></li>
<li class="toctree-l2"><a class="reference internal" href="#supported-operators">Supported operators</a></li>
</ul>
</li>
<li class="toctree-l1"><a class="reference internal" href="onnx_transfer.html">Train from ONNX models</a></li>
</ul>
<p class="caption"><span class="caption-text">Quantization and Export:</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="quant_post.html">Post-training quantization</a></li>
<li class="toctree-l1"><a class="reference internal" href="quant_qat.html">[NEW] Quantization-Aware Training</a></li>
<li class="toctree-l1"><a class="reference internal" href="export_CPP.html">Export: C++</a></li>
<li class="toctree-l1"><a class="reference internal" href="export_CPP_STM32.html">Export: C++/STM32</a></li>
<li class="toctree-l1"><a class="reference internal" href="export_TensorRT.html">Export: TensorRT</a></li>
<li class="toctree-l1"><a class="reference internal" href="export_DNeuro.html">Export: DNeuro</a></li>
<li class="toctree-l1"><a class="reference internal" href="export_ONNX.html">Export: ONNX</a></li>
<li class="toctree-l1"><a class="reference internal" href="export_legacy.html">Export: other / legacy</a></li>
</ul>
<p class="caption"><span class="caption-text">INI File Interface:</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="ini_intro.html">Introduction</a></li>
<li class="toctree-l1"><a class="reference internal" href="ini_databases.html">Databases</a></li>
<li class="toctree-l1"><a class="reference internal" href="ini_data_analysis.html">Stimuli data analysis</a></li>
<li class="toctree-l1"><a class="reference internal" href="ini_environment.html">Stimuli provider (Environment)</a></li>
<li class="toctree-l1"><a class="reference internal" href="ini_layers.html">Network Layers</a></li>
<li class="toctree-l1"><a class="reference internal" href="ini_target.html">Targets (outputs & losses)</a></li>
<li class="toctree-l1"><a class="reference internal" href="adversarial.html">Adversarial module</a></li>
</ul>
<p class="caption"><span class="caption-text">Python API:</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="containers.html">Containers</a></li>
<li class="toctree-l1"><a class="reference internal" href="cells.html">Cells</a></li>
<li class="toctree-l1"><a class="reference internal" href="databases.html">Databases</a></li>
<li class="toctree-l1"><a class="reference internal" href="stimuliprovider.html">StimuliProvider</a></li>
<li class="toctree-l1"><a class="reference internal" href="deepnet.html">DeepNet</a></li>
</ul>
<p class="caption"><span class="caption-text">C++ API / Developer:</span></p>
<ul>
<li class="toctree-l1"><a class="reference internal" href="dev_intro.html">Introduction</a></li>
</ul>
</div>
</div>
</nav>
<section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
<nav class="wy-nav-top" aria-label="top navigation">
<i data-toggle="wy-nav-top" class="fa fa-bars"></i>
<a href="index.html">N2D2</a>
</nav>
<div class="wy-nav-content">
<div class="rst-content">
<div role="navigation" aria-label="breadcrumbs navigation">
<ul class="wy-breadcrumbs">
<li><a href="index.html" class="icon icon-home"></a> »</li>
<li>Import ONNX models</li>
<li class="wy-breadcrumbs-aside">
<a href="_sources/onnx_import.rst.txt" rel="nofollow"> View page source</a>
</li>
</ul>
<hr/>
</div>
<div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
<div itemprop="articleBody">
<div class="section" id="import-onnx-models">
<h1>Import ONNX models<a class="headerlink" href="#import-onnx-models" title="Permalink to this headline">¶</a></h1>
<div class="section" id="preliminary-steps">
<h2>Preliminary steps<a class="headerlink" href="#preliminary-steps" title="Permalink to this headline">¶</a></h2>
<p>ONNX generators may generate complicated models, in order to take into account
for example dynamic size or shape calculation, from previous operator outputs
dimensions. This can be the case even when the graph is static and the dimensions
are known in the ONNX model. While such model may be imported in DL frameworks
using standard operators/layers, it would be vastly sub-optimal, as some part
of the graph would require unnecessary dynamic allocation, and would be very
hard to optimize for inference on embedded platforms.</p>
<p>For this reason, we do not always try to allow proper import of such graph in
N2D2 as is. While some simplifications may be handled directly in N2D2, we
recommend using the
<a class="reference external" href="https://github.com/daquexian/onnx-simplifier">ONNX Simplifier</a> tool on your
ONNX model before importing it into N2D2.</p>
</div>
<div class="section" id="with-an-ini-file">
<h2>With an INI file<a class="headerlink" href="#with-an-ini-file" title="Permalink to this headline">¶</a></h2>
<p>It is possible to include an ONNX model inside a N2D2 INI file, as part of a
graph. This is particularly useful to add pre-processing and post-processing to
an existing ONNX model. Below is an example with the MobileNet ONNX model
provided by Google:</p>
<div class="highlight-ini notranslate"><div class="highlight"><pre><span></span><span class="na">$BATCH_SIZE</span><span class="o">=</span><span class="s">256</span>
<span class="na">DefaultModel</span><span class="o">=</span><span class="s">Frame_CUDA</span>
<span class="c1">; Database</span>
<span class="k">[database]</span>
<span class="na">Type</span><span class="o">=</span><span class="s">ILSVRC2012_Database</span>
<span class="na">RandomPartitioning</span><span class="o">=</span><span class="s">0</span>
<span class="na">Learn</span><span class="o">=</span><span class="s">1.0</span>
<span class="na">BackgroundClass</span><span class="o">=</span><span class="s">1 ; Necessary for Google MobileNet pre-trained models</span>
<span class="c1">; Environment</span>
<span class="k">[sp]</span>
<span class="na">SizeX</span><span class="o">=</span><span class="s">224</span>
<span class="na">SizeY</span><span class="o">=</span><span class="s">224</span>
<span class="na">NbChannels</span><span class="o">=</span><span class="s">3</span>
<span class="na">BatchSize</span><span class="o">=</span><span class="s">${BATCH_SIZE}</span>
<span class="k">[sp.Transformation-1]</span>
<span class="na">Type</span><span class="o">=</span><span class="s">RescaleTransformation</span>
<span class="na">Width</span><span class="o">=</span><span class="s">256</span>
<span class="na">Height</span><span class="o">=</span><span class="s">256</span>
<span class="k">[sp.Transformation-2]</span>
<span class="na">Type</span><span class="o">=</span><span class="s">PadCropTransformation</span>
<span class="na">Width</span><span class="o">=</span><span class="s">224</span>
<span class="na">Height</span><span class="o">=</span><span class="s">224</span>
<span class="k">[sp.Transformation-3]</span>
<span class="na">Type</span><span class="o">=</span><span class="s">ColorSpaceTransformation</span>
<span class="na">ColorSpace</span><span class="o">=</span><span class="s">RGB</span>
<span class="k">[sp.Transformation-4]</span>
<span class="na">Type</span><span class="o">=</span><span class="s">RangeAffineTransformation</span>
<span class="na">FirstOperator</span><span class="o">=</span><span class="s">Minus</span>
<span class="na">FirstValue</span><span class="o">=</span><span class="s">127.5</span>
<span class="na">SecondOperator</span><span class="o">=</span><span class="s">Divides</span>
<span class="na">SecondValue</span><span class="o">=</span><span class="s">127.5</span>
<span class="c1">; Here, we insert an ONNX graph in the N2D2 flow the same way as a regular Cell</span>
<span class="k">[onnx]</span>
<span class="na">Input</span><span class="o">=</span><span class="s">sp</span>
<span class="na">Type</span><span class="o">=</span><span class="s">ONNX</span>
<span class="na">File</span><span class="o">=</span><span class="s">mobilenet_v1_1.0_224.onnx</span>
<span class="c1">; We can add targets to ONNX cells</span>
<span class="k">[MobilenetV1/Predictions/Softmax:0.Target-Top5]</span>
<span class="na">TopN</span><span class="o">=</span><span class="s">5</span>
</pre></div>
</div>
<p>A N2D2 target must be associated to the output layer of the ONNX model in order
to compute the score in N2D2.</p>
<div class="admonition note">
<p class="admonition-title">Note</p>
<p>The imported ONNX layer names in N2D2 is the name of their first output (
the operator “name” field is indeed optional in the ONNX standard).
You can easily find the ONNX cell names after running N2D2 or by opening
the ONNX graph in a graph viewer like NETRON
(<a class="reference external" href="https://lutzroeder.github.io/netron/">https://lutzroeder.github.io/netron/</a>).</p>
</div>
<p>Once the INI file including the ONNX model is ready, the following command must
be used to run N2D2 in test (inference) mode:</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">n2d2</span> <span class="n">MobileNet_ONNX</span><span class="o">.</span><span class="n">ini</span> <span class="o">-</span><span class="n">seed</span> <span class="mi">1</span> <span class="o">-</span><span class="n">w</span> <span class="o">/</span><span class="n">dev</span><span class="o">/</span><span class="n">null</span> <span class="o">-</span><span class="n">test</span>
</pre></div>
</div>
<p>There required command line arguments for running INI files including ONNX model
are described above:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 11%" />
<col style="width: 89%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Command line argument</p></th>
<th class="head"><p>Description</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">-seed</span> <span class="pre">1</span></code></p></td>
<td><p>Initial seed, necessary for test without learning before</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">-w</span> <span class="pre">/dev/null</span></code></p></td>
<td><p>No external weight loading: trained weight values are contained in the ONNX model</p></td>
</tr>
</tbody>
</table>
<div class="section" id="onnx-ini-section-type">
<h3>ONNX INI section type<a class="headerlink" href="#onnx-ini-section-type" title="Permalink to this headline">¶</a></h3>
<p>The table below summarizes the parameters of an ONNX INI section. To declare an
ONNX section, the <code class="docutils literal notranslate"><span class="pre">Type</span></code> parameter must be equal to <code class="docutils literal notranslate"><span class="pre">ONNX</span></code>. The name of the
section can be arbitrary.</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 11%" />
<col style="width: 89%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Option [default value]</p></th>
<th class="head"><p>Description</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">Type=ONNX</span></code></p></td>
<td><p>ONNX section type</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">File</span></code></p></td>
<td><p>Path to the ONNX file</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">Ignore</span></code> []</p></td>
<td><p>Space-separated list of ONNX operators to ignore during import</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">IgnoreInputSize</span></code> [0]</p></td>
<td><p>If true (1), the input size specified in the ONNX model is ignored and the N2D2 <code class="docutils literal notranslate"><span class="pre">StimuliProvider</span></code> size is used</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">Transpose</span></code> [0]</p></td>
<td><p>If true (1), the first 2 dimensions are transposed in the whole ONNX graph (1D graph are first interpreted as 2D with the second dimension equal to 1)</p></td>
</tr>
</tbody>
</table>
</div>
<div class="section" id="transpose-option-usage">
<h3><code class="docutils literal notranslate"><span class="pre">Transpose</span></code> option usage<a class="headerlink" href="#transpose-option-usage" title="Permalink to this headline">¶</a></h3>
<p>The <code class="docutils literal notranslate"><span class="pre">Transpose</span></code> option allows to transpose the first two dimensions of a whole
graph. This can be used in practice to used transposed inputs (like a transposed
image, or a transposed vector for 1D signal inputs), like shown below:</p>
<div class="highlight-ini notranslate"><div class="highlight"><pre><span></span><span class="k">[sp]</span>
<span class="na">Size</span><span class="o">=</span><span class="s">8000 1 1</span>
<span class="na">BatchSize</span><span class="o">=</span><span class="s">${BATCH_SIZE}</span>
<span class="c1">; Transpose the input:</span>
<span class="k">[trans]</span>
<span class="na">Input</span><span class="o">=</span><span class="s">sp</span>
<span class="na">Type</span><span class="o">=</span><span class="s">Transpose</span>
<span class="na">NbOutputs</span><span class="o">=</span><span class="s">1</span>
<span class="na">Perm</span><span class="o">=</span><span class="s">1 0 2 3</span>
<span class="c1">; Output dimensions are now "1 8000 1 ${BATCH_SIZE}"</span>
<span class="k">[onnx]</span>
<span class="na">Input</span><span class="o">=</span><span class="s">trans</span>
<span class="na">Type</span><span class="o">=</span><span class="s">ONNX</span>
<span class="na">Transpose</span><span class="o">=</span><span class="s">1</span>
<span class="c1">; The graph originally expects an input dimension of "8000"</span>
<span class="c1">; After "Transpose=1", the expected input dimension becomes "1 8000"</span>
<span class="na">File</span><span class="o">=</span><span class="s">sound_processing_graph.onnx</span>
</pre></div>
</div>
</div>
</div>
<div class="section" id="with-the-python-api">
<h2>With the Python API<a class="headerlink" href="#with-the-python-api" title="Permalink to this headline">¶</a></h2>
<p>The <code class="docutils literal notranslate"><span class="pre">DeepNetGenerator</span></code> can be used to load ONNX file as well as INI file.</p>
<div class="highlight-python notranslate"><div class="highlight"><pre><span></span><span class="n">net</span> <span class="o">=</span> <span class="n">N2D2</span><span class="o">.</span><span class="n">Network</span><span class="p">(</span><span class="mi">1</span><span class="p">)</span>
<span class="n">deepNet</span> <span class="o">=</span> <span class="n">N2D2</span><span class="o">.</span><span class="n">DeepNetGenerator</span><span class="o">.</span><span class="n">generate</span><span class="p">(</span><span class="n">net</span><span class="p">,</span> <span class="s2">"mobilenet_v1_1.0_224.onnx"</span><span class="p">)</span>
<span class="n">deepNet</span><span class="o">.</span><span class="n">initialize</span><span class="p">()</span>
</pre></div>
</div>
</div>
<div class="section" id="supported-operators">
<h2>Supported operators<a class="headerlink" href="#supported-operators" title="Permalink to this headline">¶</a></h2>
<p>The table below summarizes the currently implemented ONNX operators:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 29%" />
<col style="width: 14%" />
<col style="width: 57%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Operator</p></th>
<th class="head"><p>Support</p></th>
<th class="head"><p>Remarks</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td><p>Add</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>AveragePool</p></td>
<td><p>✓</p></td>
<td><p>Exc. <cite>ceil_mode</cite> and <cite>count_include_pad</cite>
attributes</p></td>
</tr>
<tr class="row-even"><td><p>BatchNormalization</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>Cast</p></td>
<td><p>✗</p></td>
<td><p>Ignored</p></td>
</tr>
<tr class="row-even"><td><p>Clip</p></td>
<td><p>✓</p></td>
<td><p>Only for <cite>min</cite> = 0 and <cite>max</cite> > 0</p></td>
</tr>
<tr class="row-odd"><td><p>Concat</p></td>
<td><p>✓</p></td>
<td><p>Only for layers that support it</p></td>
</tr>
<tr class="row-even"><td><p>Constant</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td><p>In some contexts only</p></td>
</tr>
<tr class="row-odd"><td><p>Conv</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>Dropout</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td><p>Exc. <cite>mask</cite> output</p></td>
</tr>
<tr class="row-odd"><td><p>Div</p></td>
<td><p>✓</p></td>
<td><p>With constant second operand only</p></td>
</tr>
<tr class="row-even"><td><p>Flatten</p></td>
<td><p>✓</p></td>
<td><p>Ignored (not necessary)</p></td>
</tr>
<tr class="row-odd"><td><p>Gemm</p></td>
<td><p>✓</p></td>
<td><p>Only for fully-connected layers</p></td>
</tr>
<tr class="row-even"><td><p>GlobalAveragePool</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>GlobalMaxPool</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>LRN</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>LeakyRelu</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>MatMul</p></td>
<td><p>✓</p></td>
<td><p>Only for fully-connected layers</p></td>
</tr>
<tr class="row-odd"><td><p>Max</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>MaxPool</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td><p>Exc. <cite>Indices</cite> output</p></td>
</tr>
<tr class="row-odd"><td><p>Mul</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>Pad</p></td>
<td><p>✓</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>Relu</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>Reshape</p></td>
<td><p>✓</p></td>
<td><p>Only for fixed dimensions</p></td>
</tr>
<tr class="row-odd"><td><p>Resize</p></td>
<td><p><span class="raw-html"><font color="red"></span> ✗ <span class="raw-html"></font></span></p></td>
<td><p>Planned (partially)</p></td>
</tr>
<tr class="row-even"><td><p>Shape</p></td>
<td><p>✗</p></td>
<td><p>Ignored</p></td>
</tr>
<tr class="row-odd"><td><p>Sigmoid</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>Softmax</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>Softplus</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>Squeeze</p></td>
<td><p>✗</p></td>
<td><p>Ignored</p></td>
</tr>
<tr class="row-odd"><td><p>Sub</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>Sum</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>Tanh</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-even"><td><p>Transpose</p></td>
<td><p><span class="raw-html"><font color="green"></span> ✓ <span class="raw-html"></font></span></p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p>Upsample</p></td>
<td><p><span class="raw-html"><font color="red"></span> ✗ <span class="raw-html"></font></span></p></td>
<td><p>Planned</p></td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="onnx_transfer.html" class="btn btn-neutral float-right" title="Train from ONNX models" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right"></span></a>
<a href="onnx_convert.html" class="btn btn-neutral float-left" title="Obtain ONNX models" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left"></span> Previous</a>
</div>
<hr/>
<div role="contentinfo">
<p>
© Copyright 2019, CEA LIST
</p>
</div>
Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a
<a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a>
provided by <a href="https://readthedocs.org">Read the Docs</a>.
</footer>
</div>
</div>
</section>
</div>
<script type="text/javascript">
jQuery(function () {
SphinxRtdTheme.Navigation.enable(true);
});
</script>
</body>
</html>