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
|
<!-- Copyright (c) 2020 ARM Limited. -->
<!-- -->
<!-- SPDX-License-Identifier: MIT -->
<!-- -->
<!-- HTML header for doxygen 1.8.13-->
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
<meta http-equiv="X-UA-Compatible" content="IE=9"/>
<meta name="generator" content="Doxygen 1.8.13"/>
<meta name="robots" content="NOINDEX, NOFOLLOW" />
<meta name="viewport" content="width=device-width, initial-scale=1"/>
<title>ArmNN: Parsers</title>
<link href="tabs.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="jquery.js"></script>
<script type="text/javascript" src="dynsections.js"></script>
<link href="navtree.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="resize.js"></script>
<script type="text/javascript" src="navtreedata.js"></script>
<script type="text/javascript" src="navtree.js"></script>
<script type="text/javascript">
$(document).ready(initResizable);
</script>
<link href="search/search.css" rel="stylesheet" type="text/css"/>
<script type="text/javascript" src="search/searchdata.js"></script>
<script type="text/javascript" src="search/search.js"></script>
<script type="text/x-mathjax-config">
MathJax.Hub.Config({
extensions: ["tex2jax.js"],
jax: ["input/TeX","output/HTML-CSS"],
});
</script><script type="text/javascript" src="http://cdn.mathjax.org/mathjax/latest/MathJax.js"></script>
<link href="doxygen.css" rel="stylesheet" type="text/css" />
<link href="stylesheet.css" rel="stylesheet" type="text/css"/>
</head>
<body>
<div id="top"><!-- do not remove this div, it is closed by doxygen! -->
<div id="titlearea">
<table cellspacing="0" cellpadding="0">
<tbody>
<tr style="height: 56px;">
<img alt="ArmNN" src="Arm_NN_horizontal_blue.png" style="max-width: 10rem; margin-top: .5rem; margin-left 10px"/>
<td style="padding-left: 0.5em;">
<div id="projectname">
 <span id="projectnumber">21.02</span>
</div>
</td>
</tr>
</tbody>
</table>
</div>
<!-- end header part -->
<!-- Generated by Doxygen 1.8.13 -->
<script type="text/javascript">
var searchBox = new SearchBox("searchBox", "search",false,'Search');
</script>
<script type="text/javascript" src="menudata.js"></script>
<script type="text/javascript" src="menu.js"></script>
<script type="text/javascript">
$(function() {
initMenu('',true,false,'search.php','Search');
$(document).ready(function() { init_search(); });
});
</script>
<div id="main-nav"></div>
</div><!-- top -->
<div id="side-nav" class="ui-resizable side-nav-resizable">
<div id="nav-tree">
<div id="nav-tree-contents">
<div id="nav-sync" class="sync"></div>
</div>
</div>
<div id="splitbar" style="-moz-user-select:none;"
class="ui-resizable-handle">
</div>
</div>
<script type="text/javascript">
$(document).ready(function(){initNavTree('parsers.xhtml','');});
</script>
<div id="doc-content">
<!-- window showing the filter options -->
<div id="MSearchSelectWindow"
onmouseover="return searchBox.OnSearchSelectShow()"
onmouseout="return searchBox.OnSearchSelectHide()"
onkeydown="return searchBox.OnSearchSelectKey(event)">
</div>
<!-- iframe showing the search results (closed by default) -->
<div id="MSearchResultsWindow">
<iframe src="javascript:void(0)" frameborder="0"
name="MSearchResults" id="MSearchResults">
</iframe>
</div>
<div class="header">
<div class="headertitle">
<div class="title">Parsers </div> </div>
</div><!--header-->
<div class="contents">
<div class="toc"><h3>Table of Contents</h3>
<ul><li class="level1"><a href="#S4_caffe_parser">Arm NN Caffe Parser</a></li>
<li class="level1"><a href="#S5_onnx_parser">ArmNN Onnx Parser</a></li>
<li class="level1"><a href="#S6_tf_lite_parser">ArmNN Tf Lite Parser</a></li>
<li class="level1"><a href="#S7_tf_parser">ArmNN Tensorflow Parser</a></li>
</ul>
</div>
<div class="textblock"><p>Execute models from different machine learning platforms efficiently with our parsers.</p>
<p>Simply choose a parser according to the model you want to run e.g. If you've got a model in tensorflow format (<model_name>.pb) use our tensorflow-parser.</p>
<p>If you would like to run a Tensorflow Lite (TfLite) model you probably also want to take a look at our <a class="el" href="delegate.xhtml">TfLite Delegate</a>.</p>
<p>All parsers are written in C++ but it is also possible to use them in python. For more information on our python bindings take a look into the <a class="el" href="md_python_pyarmnn__r_e_a_d_m_e.xhtml">PyArmNN</a> section.</p>
<h1><a class="anchor" id="S4_caffe_parser"></a>
Arm NN Caffe Parser</h1>
<p><code><a class="el" href="namespacearmnn_caffe_parser.xhtml" title="Caffe networks are loaded from protobuf files (binary or text) using the protobuf library and the gen...">armnnCaffeParser</a></code> is a library for loading neural networks defined in Caffe protobuf files into the Arm NN runtime.</p>
<p>Please note that certain deprecated Caffe features are not supported by the <a class="el" href="namespacearmnn_caffe_parser.xhtml" title="Caffe networks are loaded from protobuf files (binary or text) using the protobuf library and the gen...">armnnCaffeParser</a>. If you think that Arm NN should be able to load your model according to the list of supported layers, but you are getting strange error messages, then try upgrading your model to the latest format using Caffe, either by saving it to a new file or using the upgrade utilities in <code>caffe/tools</code>.</p>
<p><b>NOTE:</b> The Arm NN Caffe Parser is deprecated in Arm NN 21.02 and will be removed in 21.05.</p>
<h2>Caffe layers supported by the Arm NN SDK</h2>
<p>This reference guide provides a list of Caffe layers the Arm NN SDK currently supports.</p>
<h3>Although some other neural networks might work, Arm tests the Arm NN SDK with Caffe implementations of the following neural networks:</h3>
<ul>
<li>AlexNet.</li>
<li>Cifar10.</li>
<li>Inception-BN.</li>
<li>Resnet_50, Resnet_101 and Resnet_152.</li>
<li>VGG_CNN_S, VGG_16 and VGG_19.</li>
<li>Yolov1_tiny.</li>
<li>Lenet.</li>
<li>MobileNetv1.</li>
<li>SqueezeNet v1.0 and SqueezeNet v1.1</li>
</ul>
<h3>The Arm NN SDK supports the following machine learning layers for Caffe networks:</h3>
<ul>
<li>Argmax, excluding the top_k and out_max_val parameters.</li>
<li>BatchNorm, in inference mode.</li>
<li>Convolution, excluding Weight Filler, Bias Filler, Engine, Force nd_im2col, and Axis parameters.</li>
<li><p class="startli">Deconvolution, excluding the Dilation Size, Weight Filler, Bias Filler, Engine, Force nd_im2col, and Axis parameters.</p>
<p class="startli">Caffe doesn't support depthwise convolution, the equivalent layer is implemented through the notion of groups. ArmNN supports groups this way:</p><ul>
<li>when group=1, it is a normal conv2d</li>
<li>when group=#input_channels, we can replace it by a depthwise convolution</li>
<li>when group>1 && group<#input_channels, we need to split the input into the given number of groups, apply a separate convolution and then merge the results</li>
</ul>
</li>
<li>Concat, along the channel dimension only.</li>
<li>Dropout, in inference mode.</li>
<li>Eltwise, excluding the coeff parameter.</li>
<li>Inner Product, excluding the Weight Filler, Bias Filler, Engine, and Axis parameters.</li>
<li>Input.</li>
<li>LRN, excluding the Engine parameter.</li>
<li>Pooling, excluding the Stochastic Pooling and Engine parameters.</li>
<li>ReLU.</li>
<li>Scale.</li>
<li>Softmax, excluding the Axis and Engine parameters.</li>
<li>Split.</li>
</ul>
<p>More machine learning layers will be supported in future releases.</p>
<p><br />
<br />
<br />
<br />
</p>
<h1><a class="anchor" id="S5_onnx_parser"></a>
ArmNN Onnx Parser</h1>
<p><code><a class="el" href="namespacearmnn_onnx_parser.xhtml">armnnOnnxParser</a></code> is a library for loading neural networks defined in ONNX protobuf files into the Arm NN runtime.</p>
<h2>ONNX operators that the Arm NN SDK supports</h2>
<p>This reference guide provides a list of ONNX operators the Arm NN SDK currently supports.</p>
<p>The Arm NN SDK ONNX parser currently only supports fp32 operators.</p>
<h3>Fully supported</h3>
<ul>
<li>Add<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#Add">Add documentation</a> for more information</li>
</ul>
</li>
<li>AveragePool<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#AveragePool">AveragePool documentation</a> for more information.</li>
</ul>
</li>
<li>Constant<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#Constant">Constant documentation</a> for more information.</li>
</ul>
</li>
<li>Clip<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#Clip">Clip documentation</a> for more information.</li>
</ul>
</li>
<li>Flatten<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#Flatten">Flatten documentation</a> for more information.</li>
</ul>
</li>
<li>GlobalAveragePool<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#GlobalAveragePool">GlobalAveragePool documentation</a> for more information.</li>
</ul>
</li>
<li>LeakyRelu<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#LeakyRelu">LeakyRelu documentation</a> for more information.</li>
</ul>
</li>
<li>MaxPool<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#MaxPool">max_pool documentation</a> for more information.</li>
</ul>
</li>
<li>Relu<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#Relu">Relu documentation</a> for more information.</li>
</ul>
</li>
<li>Reshape<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#Reshape">Reshape documentation</a> for more information.</li>
</ul>
</li>
<li>Sigmoid<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#Sigmoid">Sigmoid documentation</a> for more information.</li>
</ul>
</li>
<li>Tanh<ul>
<li>See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#Tanh">Tanh documentation</a> for more information.</li>
</ul>
</li>
</ul>
<h3>Partially supported</h3>
<ul>
<li>Conv<ul>
<li>The parser only supports 2D convolutions with a group = 1 or group = #Nb_of_channel (depthwise convolution)</li>
</ul>
</li>
<li>BatchNormalization<ul>
<li>The parser does not support training mode. See the ONNX <a href="https://github.com/onnx/onnx/blob/master/docs/Operators.md#BatchNormalization">BatchNormalization documentation</a> for more information.</li>
</ul>
</li>
<li>MatMul<ul>
<li>The parser only supports constant weights in a fully connected layer.</li>
</ul>
</li>
</ul>
<h2>Tested networks</h2>
<p>Arm tested these operators with the following ONNX fp32 neural networks:</p><ul>
<li>Mobilenet_v2. See the ONNX <a href="https://github.com/onnx/models/tree/master/vision/classification/mobilenet">MobileNet documentation</a> for more information.</li>
<li>Simple MNIST. This is no longer directly documented by ONNX. The model and test data may be downloaded <a href="https://onnxzoo.blob.core.windows.net/models/opset_8/mnist/mnist.tar.gz">from the ONNX model zoo</a>.</li>
</ul>
<p>More machine learning operators will be supported in future releases. <br />
<br />
<br />
<br />
</p>
<h1><a class="anchor" id="S6_tf_lite_parser"></a>
ArmNN Tf Lite Parser</h1>
<p><code><a class="el" href="namespacearmnn_tf_lite_parser.xhtml">armnnTfLiteParser</a></code> is a library for loading neural networks defined by TensorFlow Lite FlatBuffers files into the Arm NN runtime.</p>
<h2>TensorFlow Lite operators that the Arm NN SDK supports</h2>
<p>This reference guide provides a list of TensorFlow Lite operators the Arm NN SDK currently supports.</p>
<h3>Fully supported</h3>
<p>The Arm NN SDK TensorFlow Lite parser currently supports the following operators:</p>
<ul>
<li>ADD</li>
<li>AVERAGE_POOL_2D, Supported Fused Activation: RELU , RELU6 , TANH, NONE</li>
<li>BATCH_TO_SPACE</li>
<li>CONCATENATION, Supported Fused Activation: RELU , RELU6 , TANH, NONE</li>
<li>CONV_2D, Supported Fused Activation: RELU , RELU6 , TANH, NONE</li>
<li>DEPTH_TO_SPACE</li>
<li>DEPTHWISE_CONV_2D, Supported Fused Activation: RELU , RELU6 , TANH, NONE</li>
<li>DEQUANTIZE</li>
<li>DIV</li>
<li>ELU</li>
<li>EXP</li>
<li>FULLY_CONNECTED, Supported Fused Activation: RELU , RELU6 , TANH, NONE</li>
<li>GATHER</li>
<li>HARD_SWISH</li>
<li>LEAKY_RELU</li>
<li>LOGISTIC</li>
<li>L2_NORMALIZATION</li>
<li>MAX_POOL_2D, Supported Fused Activation: RELU , RELU6 , TANH, NONE</li>
<li>MAXIMUM</li>
<li>MEAN</li>
<li>MINIMUM</li>
<li>MUL</li>
<li>NEG</li>
<li>PACK</li>
<li>PAD</li>
<li>QUANTIZE</li>
<li>RELU</li>
<li>RELU6</li>
<li>REDUCE_MAX</li>
<li>REDUCE_MIN</li>
<li>RESHAPE</li>
<li>RESIZE_BILINEAR</li>
<li>RESIZE_NEAREST_NEIGHBOR</li>
<li>SLICE</li>
<li>SOFTMAX</li>
<li>SPACE_TO_BATCH</li>
<li>SPLIT</li>
<li>SPLIT_V</li>
<li>SQUEEZE</li>
<li>STRIDED_SLICE</li>
<li>SUB</li>
<li>SUM</li>
<li>TANH</li>
<li>TRANSPOSE</li>
<li>TRANSPOSE_CONV</li>
<li>UNPACK</li>
</ul>
<h3>Custom Operator</h3>
<ul>
<li>TFLite_Detection_PostProcess</li>
</ul>
<h2>Tested networks</h2>
<p>Arm tested these operators with the following TensorFlow Lite neural network:</p><ul>
<li><a href="http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_224_quant.tgz">Quantized MobileNet</a></li>
<li><a href="http://download.tensorflow.org/models/object_detection/ssd_mobilenet_v1_quantized_300x300_coco14_sync_2018_07_18.tar.gz">Quantized SSD MobileNet</a></li>
<li>DeepSpeech v1 converted from <a href="https://github.com/mozilla/DeepSpeech/releases/tag/v0.4.1">TensorFlow model</a></li>
<li>DeepSpeaker</li>
<li><a href="https://www.tensorflow.org/lite/models/segmentation/overview">DeepLab v3+</a></li>
<li>FSRCNN</li>
<li>EfficientNet-lite</li>
<li>RDN converted from <a href="https://github.com/hengchuan/RDN-TensorFlow">TensorFlow model</a></li>
<li>Quantized RDN (CpuRef)</li>
<li><a href="http://download.tensorflow.org/models/tflite_11_05_08/inception_v3_quant.tgz">Quantized Inception v3</a></li>
<li><a href="http://download.tensorflow.org/models/inception_v4_299_quant_20181026.tgz">Quantized Inception v4</a> (CpuRef)</li>
<li>Quantized ResNet v2 50 (CpuRef)</li>
<li>Quantized Yolo v3 (CpuRef)</li>
</ul>
<p>More machine learning operators will be supported in future releases. <br />
<br />
<br />
<br />
</p>
<h1><a class="anchor" id="S7_tf_parser"></a>
ArmNN Tensorflow Parser</h1>
<p><code><a class="el" href="namespacearmnn_tf_parser.xhtml">armnnTfParser</a></code> is a library for loading neural networks defined by TensorFlow protobuf files into the Arm NN runtime.</p>
<p><b>NOTE:</b> The Arm NN Tensorflow Parser is deprecated in Arm NN 21.02 and will be removed in 21.05.</p>
<h2>TensorFlow operators that the Arm NN SDK supports</h2>
<p>This reference guide provides a list of TensorFlow operators the Arm NN SDK currently supports.</p>
<p>The Arm NN SDK TensorFlow parser currently only supports fp32 operators.</p>
<h3>Fully supported</h3>
<ul>
<li>avg_pool<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/avg_pool">avg_pool documentation</a> for more information.</li>
</ul>
</li>
<li>bias_add<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/bias_add">bias_add documentation</a> for more information.</li>
</ul>
</li>
<li>conv2d<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/conv2d">conv2d documentation</a> for more information.</li>
</ul>
</li>
<li>expand_dims<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/expand_dims">expand_dims documentation</a> for more information.</li>
</ul>
</li>
<li>gather<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/gather">gather documentation</a> for more information.</li>
</ul>
</li>
<li>identity<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/identity">identity documentation</a> for more information.</li>
</ul>
</li>
<li>local_response_normalization<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/local_response_normalization">local_response_normalization documentation</a> for more information.</li>
</ul>
</li>
<li>max_pool<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/max_pool">max_pool documentation</a> for more information.</li>
</ul>
</li>
<li>placeholder<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/placeholder">placeholder documentation</a> for more information.</li>
</ul>
</li>
<li>reduce_mean<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/reduce_mean">reduce_mean documentation</a> for more information.</li>
</ul>
</li>
<li>relu<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/relu">relu documentation</a> for more information.</li>
</ul>
</li>
<li>relu6<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/relu6">relu6 documentation</a> for more information.</li>
</ul>
</li>
<li>rsqrt<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/math/rsqrt">rsqrt documentation</a> for more information.</li>
</ul>
</li>
<li>shape<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/shape">shape documentation</a> for more information.</li>
</ul>
</li>
<li>sigmoid<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/sigmoid">sigmoid documentation</a> for more information.</li>
</ul>
</li>
<li>softplus<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/softplus">softplus documentation</a> for more information.</li>
</ul>
</li>
<li>squeeze<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/squeeze">squeeze documentation</a> for more information.</li>
</ul>
</li>
<li>tanh<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/tanh">tanh documentation</a> for more information.</li>
</ul>
</li>
<li>transpose<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/transpose">transpose documentation</a> for more information.</li>
</ul>
</li>
</ul>
<h3>Partially supported</h3>
<ul>
<li>add<ul>
<li>The parser does not support all forms of <a href="https://www.tensorflow.org/performance/xla/broadcasting">broadcast composition</a>, only broadcasting of scalars and 1D tensors. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/add">add operator documentation</a> for more information.</li>
</ul>
</li>
<li>add_n<ul>
<li>The parser does not support all forms of <a href="https://www.tensorflow.org/performance/xla/broadcasting">broadcast composition</a>, only broadcasting of scalars and 1D tensors. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/add_n">add operator documentation</a> for more information.</li>
</ul>
</li>
<li>concat<ul>
<li>Arm NN supports concatenation along the channel dimension for data formats NHWC and NCHW.</li>
</ul>
</li>
<li>constant<ul>
<li>The parser does not support the optional <code>shape</code> argument. It always infers the shape of the output tensor from <code>value</code>. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/constant">constant documentation</a> for further information.</li>
</ul>
</li>
<li>depthwise_conv2d_native<ul>
<li>The parser only supports a dilation rate of (1,1,1,1). See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/depthwise_conv2d_native">depthwise_conv2d_native documentation</a> for more information.</li>
</ul>
</li>
<li>equal<ul>
<li>The parser does not support all forms of <a href="https://www.tensorflow.org/performance/xla/broadcasting">broadcast composition</a>, only broadcasting of 4D and 1D tensors. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/math/equal">equal operator documentation</a> for more information.</li>
</ul>
</li>
<li>fused_batch_norm<ul>
<li>The parser does not support training outputs. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/fused_batch_norm">fused_batch_norm documentation</a> for more information.</li>
</ul>
</li>
<li>greater<ul>
<li>The parser does not support all forms of <a href="https://www.tensorflow.org/performance/xla/broadcasting">broadcast composition</a>, only broadcasting of 4D and 1D tensors. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/math/greater">greater operator documentation</a> for more information.</li>
</ul>
</li>
<li>matmul<ul>
<li>The parser only supports constant weights in a fully connected layer. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/matmul">matmul documentation</a> for more information.</li>
</ul>
</li>
<li>maximum where maximum is used in one of the following ways<ul>
<li>max(mul(a, x), x)</li>
<li>max(mul(x, a), x)</li>
<li>max(x, mul(a, x))</li>
<li>max(x, mul(x, a) This is interpreted as a <a class="el" href="classarmnn_1_1_activation_layer.xhtml" title="This layer represents an activation operation with the specified activation function. ">ActivationLayer</a> with a LeakyRelu activation function. Any other usage of max will result in the insertion of a simple maximum layer. The parser does not support all forms of <a href="https://www.tensorflow.org/performance/xla/broadcasting">broadcast composition</a>. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/maximum">maximum documentation</a> for more information.</li>
</ul>
</li>
<li>minimum<ul>
<li>The parser does not support all forms of <a href="https://www.tensorflow.org/performance/xla/broadcasting">broadcast composition</a>, only broadcasting of 4D and 1D tensors. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/math/minimum">minimum operator documentation</a> for more information.</li>
</ul>
</li>
<li>multiply<ul>
<li>The parser does not support all forms of <a href="https://www.tensorflow.org/performance/xla/broadcasting">broadcast composition</a>, only broadcasting of scalars and 1D tensors. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/multiply">multiply documentation</a> for more information.</li>
</ul>
</li>
<li>pack/stack<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/stack">stack documentation</a> for more information.</li>
</ul>
</li>
<li>pad<ul>
<li>Only supports tf.pad function with mode = 'CONSTANT' and constant_values = 0. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/pad">pad documentation</a> for more information.</li>
</ul>
</li>
<li>realdiv<ul>
<li>The parser does not support all forms of <a href="https://www.tensorflow.org/performance/xla/broadcasting">broadcast composition</a>, only broadcasting of scalars and 1D tensors. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/realdiv">realdiv documentation</a> for more information.</li>
</ul>
</li>
<li>reshape<ul>
<li>The parser does not support reshaping to or from 4D. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/reshape">reshape documentation</a> for more information.</li>
</ul>
</li>
<li>resize_images<ul>
<li>The parser only supports <code>ResizeMethod.BILINEAR</code> with <code>align_corners=False</code>. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/image/resize_images">resize_images documentation</a> for more information.</li>
</ul>
</li>
<li>softmax<ul>
<li>The parser only supports 2D inputs and does not support selecting the <code>softmax</code> dimension. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/nn/softmax">softmax documentation</a> for more information.</li>
</ul>
</li>
<li>split<ul>
<li>Arm NN supports split along the channel dimension for data formats NHWC and NCHW.</li>
</ul>
</li>
<li>strided_slice<ul>
<li>See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/strided_slice">strided_slice documentation</a> for more information.</li>
</ul>
</li>
<li>subtract<ul>
<li>The parser does not support all forms of broadcasting <a href="https://www.tensorflow.org/performance/xla/broadcasting">broadcast composition</a>, only broadcasting of scalars and 1D tensors. See the TensorFlow <a href="https://www.tensorflow.org/api_docs/python/tf/math/subtract">subtract documentation</a> for more information.</li>
</ul>
</li>
</ul>
<h2>Tested networks</h2>
<p>Arm tests these operators with the following TensorFlow fp32 neural networks:</p><ul>
<li>Lenet</li>
<li>mobilenet_v1_1.0_224. The Arm NN SDK only supports the non-quantized version of the network. See the https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet_v1.md "MobileNet_v1 documentation" for more information on quantized networks.</li>
<li>inception_v3. The Arm NN SDK only supports the official inception_v3 transformed model. See the TensorFlow documentation on <a href="https://www.tensorflow.org/mobile/prepare_models">preparing models for mobile deployment</a> for more information on how to transform the inception_v3 network.</li>
<li>Simple MNIST. For more information check out the <a href="https://developer.arm.com/technologies/machine-learning-on-arm/developer-material/how-to-guides/deploying-a-tensorflow-mnist-model-on-arm-nn">tutorial</a> on the Arm Developer portal.</li>
<li>ResNet v2 50 implementation from the <a href="https://github.com/tensorflow/models/tree/master/research/slim">TF Slim model zoo</a></li>
</ul>
<p>More machine learning operators will be supported in future releases. </p>
</div></div><!-- contents -->
</div><!-- doc-content -->
<!-- start footer part -->
<div id="nav-path" class="navpath"><!-- id is needed for treeview function! -->
<ul>
<li class="navelem"><a class="el" href="swtools.xhtml">Software Tools</a></li>
<li class="footer">Generated on Thu Feb 25 2021 17:27:56 for ArmNN by
<a href="http://www.doxygen.org/index.html">
<img class="footer" src="doxygen.png" alt="doxygen"/></a> 1.8.13 </li>
</ul>
</div>
</body>
</html>
|
