-
Notifications
You must be signed in to change notification settings - Fork 1
/
export_TensorRT.html
639 lines (521 loc) · 34.2 KB
/
export_TensorRT.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
<!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>Export: TensorRT — 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="Export: DNeuro" href="export_DNeuro.html" />
<link rel="prev" title="Export: C++/STM32" href="export_CPP_STM32.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>
<li class="toctree-l1"><a class="reference internal" href="onnx_convert.html">Obtain ONNX models</a></li>
<li class="toctree-l1"><a class="reference internal" href="onnx_import.html">Import ONNX models</a></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 class="current">
<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 current"><a class="current reference internal" href="#">Export: TensorRT</a><ul>
<li class="toctree-l2"><a class="reference internal" href="#informations">Informations</a></li>
<li class="toctree-l2"><a class="reference internal" href="#export-parameters">Export parameters</a></li>
<li class="toctree-l2"><a class="reference internal" href="#benchmark-your-tensorrt-model-c-benchmark">Benchmark your TensorRT Model - C++ Benchmark</a><ul>
<li class="toctree-l3"><a class="reference internal" href="#analyse-the-execution-performances-of-your-tensorrt-model-fp32">Analyse the execution performances of your TensorRT Model (FP32)</a></li>
</ul>
</li>
<li class="toctree-l2"><a class="reference internal" href="#deploy-your-tensorrt-model-in-application">Deploy your TensorRT Model in Application</a></li>
</ul>
</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>Export: TensorRT</li>
<li class="wy-breadcrumbs-aside">
<a href="_sources/export_TensorRT.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="export-tensorrt">
<h1>Export: TensorRT<a class="headerlink" href="#export-tensorrt" title="Permalink to this headline">¶</a></h1>
<p>Export type: <code class="docutils literal notranslate"><span class="pre">CPP_TensorRT</span></code></p>
<p>C++ export using TensorRT.</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">export</span> <span class="n">CPP_TensorRT</span> <span class="o">-</span><span class="n">nbbits</span> <span class="o">-</span><span class="mi">32</span>
</pre></div>
</div>
<div class="admonition warning">
<p class="admonition-title">Warning</p>
<p>The calibration for this export is done using the tools provided by NVIDIA. For this reason, you cannot calibrate when exporting your network.
You need to use the export to calibrate your network.</p>
</div>
<div class="section" id="informations">
<h2>Informations<a class="headerlink" href="#informations" title="Permalink to this headline">¶</a></h2>
<p>In order to exploit TensorRT optimizations, the N2D2 Framework provide a code generator linked with a C++/Python API
that gives access to TensorRT methods, I/O handling and specifics control.
The generated code is provided as a standalone code with its own compilation environment under a Makefile format. Moreover
a benchmark environment with stimuli from the test dataset is given to evaluates execution time performances of your model.</p>
<p>This allow a low level of dependency, only TensorRT, CUDA, cuDNN, cuBLAS and GCC are needed.
We recommended you to ensure the correct compatibility of your installation by referring to the TensorRT archive page:
<a class="reference external" href="https://docs.nvidia.com/deeplearning/tensorrt/archives/index.html">https://docs.nvidia.com/deeplearning/tensorrt/archives/index.html</a>
Follow the support matrix section of your TensorRT version, notice that TensorRT export
have been tested from TensorRT 2.1 to TensorRT 7.2 versions.</p>
<p>The TensorRT library includes implementation for the most common deep learning layers, but strong limitations
are known depending of the TensorRT version. For example, TensorRT provide a support to the well-known resize layer since version 6.0.1.
This layer is widely use for decoder, segmentation and detector tasks. For inferior version a support have been integrated under a plugin
layer. The TensorRT plugin layers allow the application to implement not supported layers.
You can find additional informations about how to implements new plugin layers here :
<a class="reference external" href="https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#add_custom_layer">https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#add_custom_layer</a></p>
<p>The plugin layers that N2D2 TensorRT generator implements are available in the folder <code class="docutils literal notranslate"><span class="pre">export/CPP_TensorRT/include/plugins/</span></code>. These layers are
used by the N2D2 TensorRT generator when TensorRT doesn’t provide support to a requested layers.</p>
</div>
<div class="section" id="export-parameters">
<h2>Export parameters<a class="headerlink" href="#export-parameters" title="Permalink to this headline">¶</a></h2>
<p>Extra parameters can be passed during export using the
<code class="docutils literal notranslate"><span class="pre">-export-parameters</span> <span class="pre">params.ini</span></code> command line argument. The parameters must be
saved in an INI-like file.</p>
<p>List of available parameters:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 35%" />
<col style="width: 65%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Argument [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">GenStimuliCalib</span></code> [1]</p></td>
<td><p>If true (1), generate calibration files, necessary for 8-bits precision.
Beware that calibration files may take a lot of disk space!</p></td>
</tr>
</tbody>
</table>
</div>
<div class="section" id="benchmark-your-tensorrt-model-c-benchmark">
<h2>Benchmark your TensorRT Model - C++ Benchmark<a class="headerlink" href="#benchmark-your-tensorrt-model-c-benchmark" title="Permalink to this headline">¶</a></h2>
<p>The TensorRT export is given with a C++ benchmark ready to be used. The benchmark program is able to evaluates the applicative performances
of your model on the test dataset exported under the <code class="docutils literal notranslate"><span class="pre">stimuli</span></code> folder at export time. A per-layer execution time analysis is also performed
to evaluates your mode latency and identify potential bottleneck.
Moreover different numerical precision supporter by NVIDIA GPU can be evaluates in order to assess potential acceleration factor and eventual
applicative performances losses.</p>
<p>When numerical precision is set to <code class="docutils literal notranslate"><span class="pre">8</span></code> bits for benchmark, the program will use the calibration files exported under the <code class="docutils literal notranslate"><span class="pre">batches_calib</span></code> folder
at export time. The calibration files also corresponds to the test stimuli pre-processes for the <code class="docutils literal notranslate"><span class="pre">IInt8EntropyCalibrator2</span></code> that implement TensorRT.
You can find more informations about the INT8 calibration procedure with TensorRT here :
<a class="reference external" href="https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#optimizing_int8_c">https://docs.nvidia.com/deeplearning/tensorrt/developer-guide/index.html#optimizing_int8_c</a></p>
<p>The command to compile and execute the C++ TensorRT Benchmark under a <code class="docutils literal notranslate"><span class="pre">FP32</span></code> precision is :</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">make</span>
<span class="n">cd</span> <span class="n">export_CPP_TensorRT_float32</span><span class="o">/</span>
<span class="o">./</span><span class="nb">bin</span><span class="o">/</span><span class="n">n2d2_tensorRT_test</span> <span class="o">-</span><span class="n">nbbits</span> <span class="o">-</span><span class="mi">32</span>
</pre></div>
</div>
<p>To launch the Benchmark in <code class="docutils literal notranslate"><span class="pre">FP16</span></code> (half precision) use this command :</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="o">./</span><span class="nb">bin</span><span class="o">/</span><span class="n">n2d2_tensorRT_test</span> <span class="o">-</span><span class="n">nbbits</span> <span class="o">-</span><span class="mi">16</span>
</pre></div>
</div>
<p>To launch the Benchmark in <code class="docutils literal notranslate"><span class="pre">INT8</span></code> use this command :</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="o">./</span><span class="nb">bin</span><span class="o">/</span><span class="n">n2d2_tensorRT_test</span> <span class="o">-</span><span class="n">nbbits</span> <span class="mi">8</span>
</pre></div>
</div>
<p>List of the program option related to the TensorRT C++ benchmark:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 7%" />
<col style="width: 93%" />
</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">-batch</span></code> [1]</p></td>
<td><p>Size of the batch to use</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">-dev</span></code> [0]</p></td>
<td><p>CUDA Device ID selection</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">-stimulus</span></code> [NULL]</p></td>
<td><p>Path to a specific input stimulus to test. For example: -stimulus <span class="math notranslate nohighlight">\({/stimulus/env0000.pgm}\)</span> command will test the file env0000.pgm of the stimulus folder.</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">-prof</span></code></p></td>
<td><p>Activates the layer wise profiling mechanism. This option can decrease execution time performance.</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">-iter-build</span></code> [1]</p></td>
<td><p>Sets the number of minimization build iterations done by the tensorRT builder to find the best layer tactics.</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">-nbbits</span></code> [-32]</p></td>
<td><p>Number of bits used for computation. Value -32 for Full FP32 bits configuration, -16 for Half FP16 bits configuration and 8 for INT8 bits configuration. When running INT8 mode for the first time, the TensorRT calibration process can be very long. Once generated the generated calibration table will be automatically reused. Supported compute mode in function of the compute capability are provided here: <a class="reference external" href="https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#compute-capabilities">https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#compute-capabilities</a> .</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">-calib-cache</span></code></p></td>
<td><p>Path and name to the calibration file generated by TensorRT calibrator when precision is INT8. Must be compatible with the TensorRT Entropy Calibrator version used to calibrate.</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">-calib-folder</span></code> [batches_calib]</p></td>
<td><p>Path to the calibration data samples. This is mandatory when precision is set to INT8 and if no calibration file cache is load.</p></td>
</tr>
</tbody>
</table>
<div class="section" id="analyse-the-execution-performances-of-your-tensorrt-model-fp32">
<h3>Analyse the execution performances of your TensorRT Model (FP32)<a class="headerlink" href="#analyse-the-execution-performances-of-your-tensorrt-model-fp32" title="Permalink to this headline">¶</a></h3>
<p>Here is a small example that described how to report the per-layer analysis on execution time.</p>
<p>Launch the Benchmark with the <code class="docutils literal notranslate"><span class="pre">-prof</span></code> argument :</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="o">./</span><span class="nb">bin</span><span class="o">/</span><span class="n">n2d2_tensorRT_test</span> <span class="o">-</span><span class="n">prof</span>
</pre></div>
</div>
<p>At the end of the execution the performances analysis is displayed in your screen :</p>
<div class="highlight-console notranslate"><div class="highlight"><pre><span></span><span class="gp gp-VirtualEnv">(19%)</span> <span class="go">**************************************** CONV1 + CONV1_ACTIVATION: 0.0219467 ms</span>
<span class="gp gp-VirtualEnv">(05%)</span> <span class="go">************ POOL1: 0.00675573 ms</span>
<span class="gp gp-VirtualEnv">(13%)</span> <span class="go">**************************** CONV2 + CONV2_ACTIVATION: 0.0159089 ms</span>
<span class="gp gp-VirtualEnv">(05%)</span> <span class="go">************ POOL2: 0.00616047 ms</span>
<span class="gp gp-VirtualEnv">(14%)</span> <span class="go">****************************** CONV3 + CONV3_ACTIVATION: 0.0159713 ms</span>
<span class="gp gp-VirtualEnv">(19%)</span> <span class="go">**************************************** FC1 + FC1_ACTIVATION: 0.0222242 ms</span>
<span class="gp gp-VirtualEnv">(13%)</span> <span class="go">**************************** FC2: 0.0149013 ms</span>
<span class="gp gp-VirtualEnv">(08%)</span> <span class="go">****************** SOFTMAX: 0.0100633 ms</span>
<span class="go">Average profiled tensorRT process time per stimulus = 0.113932 ms</span>
</pre></div>
</div>
<p>You can evaluates impact of the performances for various batch size and the different numerical precision supported.</p>
</div>
</div>
<div class="section" id="deploy-your-tensorrt-model-in-application">
<h2>Deploy your TensorRT Model in Application<a class="headerlink" href="#deploy-your-tensorrt-model-in-application" title="Permalink to this headline">¶</a></h2>
<p>The TensorRT export is provided with a C++ and a python interface. The python interface is accessible through a wrapper
to the C++ API method and linked with the libboost-python librabry.</p>
<p>You can integrates your model in your application environment as a library thanks to this API.</p>
<p>The command to compile the TensorRT export as a C++ library is :</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">make</span> <span class="n">WRAPPER_CPP</span><span class="o">=</span><span class="mi">1</span>
</pre></div>
</div>
<p>The library of your TensorRT model is provided under the name <code class="docutils literal notranslate"><span class="pre">libn2d2_tensorRT_inference.so</span></code> locate at <code class="docutils literal notranslate"><span class="pre">bin/</span></code> folder.</p>
<p>The command to compile your TensorRT export as a Python3.6m library is :</p>
<div class="highlight-default notranslate"><div class="highlight"><pre><span></span><span class="n">make</span> <span class="n">WRAPPER_PYTHON</span><span class="o">=</span><span class="mf">3.6</span><span class="n">m</span>
</pre></div>
</div>
<p>The python library of your TensorRT model is then provided under the name <code class="docutils literal notranslate"><span class="pre">N2D2.so</span></code> locate at <code class="docutils literal notranslate"><span class="pre">bin/</span></code> folder.</p>
<p>Method accessible through C++ or Python API are listed and detailled here:</p>
<table class="docutils align-default">
<colgroup>
<col style="width: 3%" />
<col style="width: 6%" />
<col style="width: 6%" />
<col style="width: 6%" />
<col style="width: 13%" />
<col style="width: 54%" />
<col style="width: 13%" />
</colgroup>
<thead>
<tr class="row-odd"><th class="head"><p>Return Type</p></th>
<th class="head"><p>C++ API</p></th>
<th class="head"><p>Python API</p></th>
<th class="head"><p>Args Type</p></th>
<th class="head"><p>ArgsName(Default Value)</p></th>
<th class="head"><p>Description</p></th>
<th class="head"><p>Comments</p></th>
</tr>
</thead>
<tbody>
<tr class="row-even"><td></td>
<td><p>N2D2::Network()</p></td>
<td><p>N2D2_Network()</p></td>
<td></td>
<td></td>
<td><p>TensorRT DNN object creation</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setMaxBatchSize</p></td>
<td><p>setMaxBatchSize</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">size_t</span></code></p></td>
<td><p>batchsize(<code class="docutils literal notranslate"><span class="pre">1</span></code>)</p></td>
<td><p>Maximum batchsize for setting the internal tensorrt graph memory usage limit</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setDeviceID</p></td>
<td><p>setDeviceID</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">size_t</span></code></p></td>
<td><p>device(<code class="docutils literal notranslate"><span class="pre">0</span></code>)</p></td>
<td><p>Device ID on which run the TensorRT model</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setPrecision</p></td>
<td><p>setPrecision</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">int</span></code></p></td>
<td><p>precision(<code class="docutils literal notranslate"><span class="pre">-32</span></code>)</p></td>
<td><p>Numerical Precision to use: <code class="docutils literal notranslate"><span class="pre">-32</span></code> for float, <code class="docutils literal notranslate"><span class="pre">-16</span></code> for half float, <code class="docutils literal notranslate"><span class="pre">8</span></code> for int8</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setDetectorThresholds</p></td>
<td><p>setDetectorThresholds</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">float*</span></code>, <code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>thresholds, lengthThreshold</p></td>
<td><p>Set the confidences thresholds of a detector output. Bypass the internal thresholds from the exported model</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setDetectorNMS</p></td>
<td><p>setDetectorNMS</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">double</span></code></p></td>
<td><p>thresholdNms</p></td>
<td><p>Set the threshold for non-maxima suppression range (from 0.0 to 1.0) of a detector output. Bypass the internal thresholds from the exported model</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setInputEngine</p></td>
<td><p>setInputEngine</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">string</span></code></p></td>
<td><p>enginePath</p></td>
<td><p>Path of a serialized and optimized TensorRT plan file. The serialized plan file are not portable across platforms or TensorRT versions and are specific to the exact GPU model they were built on</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setOutputEngine</p></td>
<td><p>setOutputEngine</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">string</span></code></p></td>
<td><p>enginePath</p></td>
<td><p>Path and name of where to save your serialized plan file. The serialized plan file are not portable across platforms or TensorRT versions and are specific to the exact GPU model they were built on</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setCalibCache</p></td>
<td><p>setCalibCache</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">string</span></code></p></td>
<td><p>calibPath</p></td>
<td><p>Path and name to the calibration file generated by TensorRT calibrator when precision is INT8. Must be compatible with the TensorRT version used to calibrate.</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setCalibFolder</p></td>
<td><p>setCalibFolder</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">string</span></code></p></td>
<td><p>folderPath(<code class="docutils literal notranslate"><span class="pre">batches_calib/</span></code>)</p></td>
<td><p>Path to the calibration data samples. This is mandatory when precision is set to INT8 and if no calibration file cache is load.</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setParamPath</p></td>
<td><p>setParamPath</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">string</span></code></p></td>
<td><p>parmPath(<code class="docutils literal notranslate"><span class="pre">dnn/</span></code>)</p></td>
<td><p>Path to the DNN parameters generated by N2D2</p></td>
<td><p>Use before run initialize()</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>initialize</p></td>
<td><p>initialize</p></td>
<td></td>
<td></td>
<td><p>Initialize the TensorRT engine following the specified options. This function allocates the memory workspace I/O, set the precision, create the builder, create the network topology from the Network generated with N2D2 and initialize the TensorRT context</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>setProfiling</p></td>
<td><p>setProfiling</p></td>
<td></td>
<td></td>
<td><p>Initialize the profiler in the TensorRT engine</p></td>
<td><p>Use after run initialize()</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>getInputDimX</p></td>
<td><p>getInputDimX</p></td>
<td></td>
<td></td>
<td><p>Return the dimension X of the input</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>getInputDimY</p></td>
<td><p>getInputDimY</p></td>
<td></td>
<td></td>
<td><p>Return the dimension Y of the input</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>getInputDimZ</p></td>
<td><p>getInputDimZ</p></td>
<td></td>
<td></td>
<td><p>Return the dimension Z of the input</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>getOutputNbTargets</p></td>
<td><p>getOutputNbTargets</p></td>
<td></td>
<td></td>
<td><p>Return the number of outputs</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>getOutputDimX</p></td>
<td><p>getOutputDimX</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>outputNumber</p></td>
<td><p>Return the dimension X of a specified output</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>getOutputDimY</p></td>
<td><p>getOutputDimY</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>outputNumber</p></td>
<td><p>Return the dimension Y of a specified output</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>getOutputDimZ</p></td>
<td><p>getOutputDimZ</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>outputNumber</p></td>
<td><p>Return the dimension Z of a specified output</p></td>
<td></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>getOutputTarget</p></td>
<td><p>getOutputTarget</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>outputNumber</p></td>
<td><p>Return the number of labels if a specified output</p></td>
<td></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>syncExe</p></td>
<td><p>syncExe</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">float*</span></code>, <code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>inputData,batchsize</p></td>
<td><p>Synchronously execute inference on a batch of the specified size</p></td>
<td><p>Use after run initialize()</p></td>
</tr>
<tr class="row-even"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>log_output</p></td>
<td><p>cpyOutput</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">float*</span></code>, <code class="docutils literal notranslate"><span class="pre">uint</span></code></p></td>
<td><p>inputData,outputID</p></td>
<td><p>Copy the full batch of the output of a DNN</p></td>
<td><p>Use after run initialize()</p></td>
</tr>
<tr class="row-odd"><td><p><code class="docutils literal notranslate"><span class="pre">void</span></code></p></td>
<td><p>estimated</p></td>
<td><p>estimated</p></td>
<td><p><code class="docutils literal notranslate"><span class="pre">float*</span></code>, <code class="docutils literal notranslate"><span class="pre">uint</span></code>,
<code class="docutils literal notranslate"><span class="pre">bool</span></code>, <code class="docutils literal notranslate"><span class="pre">float</span></code></p></td>
<td><p>inputData,outputID, useGPU, threshold</p></td>
<td><p>Copy per output pixel estimated labels of a specified output. UseGpu is recommended and threshold value allow to clip the outputs values before classification</p></td>
<td><p>Use after run initialize()</p></td>
</tr>
</tbody>
</table>
</div>
</div>
</div>
</div>
<footer>
<div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
<a href="export_DNeuro.html" class="btn btn-neutral float-right" title="Export: DNeuro" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right"></span></a>
<a href="export_CPP_STM32.html" class="btn btn-neutral float-left" title="Export: C++/STM32" 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>