xref: /aosp_15_r20/external/ComputeLibrary/examples/graph_mobilenet.cpp (revision c217d954acce2dbc11938adb493fc0abd69584f3)
1 /*
2  * Copyright (c) 2017-2021 Arm Limited.
3  *
4  * SPDX-License-Identifier: MIT
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a copy
7  * of this software and associated documentation files (the "Software"), to
8  * deal in the Software without restriction, including without limitation the
9  * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
10  * sell copies of the Software, and to permit persons to whom the Software is
11  * furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in all
14  * copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
19  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
21  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22  * SOFTWARE.
23  */
24 #include "arm_compute/graph.h"
25 #include "support/ToolchainSupport.h"
26 #include "utils/CommonGraphOptions.h"
27 #include "utils/GraphUtils.h"
28 #include "utils/Utils.h"
29 
30 using namespace arm_compute;
31 using namespace arm_compute::utils;
32 using namespace arm_compute::graph::frontend;
33 using namespace arm_compute::graph_utils;
34 
35 /** Example demonstrating how to implement MobileNet's network using the Compute Library's graph API */
36 class GraphMobilenetExample : public Example
37 {
38 public:
GraphMobilenetExample()39     GraphMobilenetExample()
40         : cmd_parser(), common_opts(cmd_parser), common_params(), graph(0, "MobileNetV1")
41     {
42         // Add model id option
43         model_id_opt = cmd_parser.add_option<SimpleOption<int>>("model-id", 0);
44         model_id_opt->set_help("Mobilenet model id (0: 1.0_224, else: 0.75_160");
45     }
46     GraphMobilenetExample(const GraphMobilenetExample &) = delete;
47     GraphMobilenetExample &operator=(const GraphMobilenetExample &) = delete;
48     ~GraphMobilenetExample() override                               = default;
do_setup(int argc,char ** argv)49     bool do_setup(int argc, char **argv) override
50     {
51         // Parse arguments
52         cmd_parser.parse(argc, argv);
53         cmd_parser.validate();
54 
55         // Consume common parameters
56         common_params = consume_common_graph_parameters(common_opts);
57 
58         // Return when help menu is requested
59         if(common_params.help)
60         {
61             cmd_parser.print_help(argv[0]);
62             return false;
63         }
64 
65         // Print parameter values
66         std::cout << common_params << std::endl;
67 
68         // Get model parameters
69         int model_id = model_id_opt->value();
70 
71         // Create input descriptor
72         unsigned int spatial_size = (model_id == 0 || common_params.data_type == DataType::QASYMM8) ? 224 : 160;
73 
74         // Create input descriptor
75         const TensorShape tensor_shape     = permute_shape(TensorShape(spatial_size, spatial_size, 3U, common_params.batches), DataLayout::NCHW, common_params.data_layout);
76         TensorDescriptor  input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
77 
78         // Set graph hints
79         graph << common_params.target
80               << common_params.fast_math_hint;
81 
82         // Create core graph
83         if(arm_compute::is_data_type_float(common_params.data_type))
84         {
85             create_graph_float(input_descriptor, model_id);
86         }
87         else
88         {
89             create_graph_qasymm(input_descriptor);
90         }
91 
92         // Create common tail
93         graph << ReshapeLayer(TensorShape(1001U)).set_name("Reshape")
94               << SoftmaxLayer().set_name("Softmax")
95               << OutputLayer(get_output_accessor(common_params, 5));
96 
97         // Finalize graph
98         GraphConfig config;
99         config.num_threads = common_params.threads;
100         config.use_tuner   = common_params.enable_tuner;
101         config.tuner_mode  = common_params.tuner_mode;
102         config.tuner_file  = common_params.tuner_file;
103         config.mlgo_file   = common_params.mlgo_file;
104 
105         graph.finalize(common_params.target, config);
106 
107         return true;
108     }
do_run()109     void do_run() override
110     {
111         // Run graph
112         graph.run();
113     }
114 
115 private:
116     CommandLineParser  cmd_parser;
117     CommonGraphOptions common_opts;
118     SimpleOption<int> *model_id_opt{ nullptr };
119     CommonGraphParams  common_params;
120     Stream             graph;
121 
create_graph_float(TensorDescriptor & input_descriptor,int model_id)122     void create_graph_float(TensorDescriptor &input_descriptor, int model_id)
123     {
124         float       depth_scale = (model_id == 0) ? 1.f : 0.75;
125         std::string model_path  = (model_id == 0) ? "/cnn_data/mobilenet_v1_1_224_model/" : "/cnn_data/mobilenet_v1_075_160_model/";
126 
127         // Create a preprocessor object
128         std::unique_ptr<IPreprocessor> preprocessor = std::make_unique<TFPreproccessor>();
129 
130         // Get trainable parameters data path
131         std::string data_path = common_params.data_path;
132 
133         // Add model path to data path
134         if(!data_path.empty())
135         {
136             data_path += model_path;
137         }
138 
139         graph << InputLayer(input_descriptor,
140                             get_input_accessor(common_params, std::move(preprocessor), false))
141               << ConvolutionLayer(
142                   3U, 3U, 32U * depth_scale,
143                   get_weights_accessor(data_path, "Conv2d_0_weights.npy", DataLayout::NCHW),
144                   std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
145                   PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::FLOOR))
146               .set_name("Conv2d_0")
147               << BatchNormalizationLayer(
148                   get_weights_accessor(data_path, "Conv2d_0_BatchNorm_moving_mean.npy"),
149                   get_weights_accessor(data_path, "Conv2d_0_BatchNorm_moving_variance.npy"),
150                   get_weights_accessor(data_path, "Conv2d_0_BatchNorm_gamma.npy"),
151                   get_weights_accessor(data_path, "Conv2d_0_BatchNorm_beta.npy"),
152                   0.001f)
153               .set_name("Conv2d_0/BatchNorm")
154               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.f)).set_name("Conv2d_0/Relu6");
155         graph << get_dwsc_node_float(data_path, "Conv2d_1", 64 * depth_scale, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0));
156         graph << get_dwsc_node_float(data_path, "Conv2d_2", 128 * depth_scale, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
157         graph << get_dwsc_node_float(data_path, "Conv2d_3", 128 * depth_scale, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
158         graph << get_dwsc_node_float(data_path, "Conv2d_4", 256 * depth_scale, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
159         graph << get_dwsc_node_float(data_path, "Conv2d_5", 256 * depth_scale, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
160         graph << get_dwsc_node_float(data_path, "Conv2d_6", 512 * depth_scale, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
161         graph << get_dwsc_node_float(data_path, "Conv2d_7", 512 * depth_scale, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
162         graph << get_dwsc_node_float(data_path, "Conv2d_8", 512 * depth_scale, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
163         graph << get_dwsc_node_float(data_path, "Conv2d_9", 512 * depth_scale, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
164         graph << get_dwsc_node_float(data_path, "Conv2d_10", 512 * depth_scale, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
165         graph << get_dwsc_node_float(data_path, "Conv2d_11", 512 * depth_scale, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
166         graph << get_dwsc_node_float(data_path, "Conv2d_12", 1024 * depth_scale, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
167         graph << get_dwsc_node_float(data_path, "Conv2d_13", 1024 * depth_scale, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::CEIL), PadStrideInfo(1, 1, 0, 0));
168         graph << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, common_params.data_layout)).set_name("Logits/AvgPool_1a")
169               << ConvolutionLayer(
170                   1U, 1U, 1001U,
171                   get_weights_accessor(data_path, "Logits_Conv2d_1c_1x1_weights.npy", DataLayout::NCHW),
172                   get_weights_accessor(data_path, "Logits_Conv2d_1c_1x1_biases.npy"),
173                   PadStrideInfo(1, 1, 0, 0))
174               .set_name("Logits/Conv2d_1c_1x1");
175     }
176 
create_graph_qasymm(TensorDescriptor & input_descriptor)177     void create_graph_qasymm(TensorDescriptor &input_descriptor)
178     {
179         // Get trainable parameters data path
180         std::string data_path = common_params.data_path;
181 
182         // Add model path to data path
183         if(!data_path.empty())
184         {
185             data_path += "/cnn_data/mobilenet_qasymm8_model/";
186         }
187 
188         // Quantization info taken from the AndroidNN QASYMM8 MobileNet example
189         const QuantizationInfo in_quant_info = QuantizationInfo(0.0078125f, 128);
190 
191         const std::vector<QuantizationInfo> conv_weights_quant_info =
192         {
193             QuantizationInfo(0.02182667888700962f, 151), // conv0
194             QuantizationInfo(0.004986600950360298f, 74)  // conv14
195         };
196         const std::vector<QuantizationInfo> conv_out_quant_info =
197         {
198             QuantizationInfo(0.023528477177023888f, 0), // conv0
199             QuantizationInfo(0.16609922051429749f, 66)  // conv14
200         };
201 
202         const std::vector<QuantizationInfo> depth_weights_quant_info =
203         {
204             QuantizationInfo(0.29219913482666016f, 110),  // dwsc1
205             QuantizationInfo(0.40277284383773804f, 130),  // dwsc2
206             QuantizationInfo(0.06053730100393295f, 160),  // dwsc3
207             QuantizationInfo(0.01675807684659958f, 123),  // dwsc4
208             QuantizationInfo(0.04105526953935623f, 129),  // dwsc5
209             QuantizationInfo(0.013460792601108551f, 122), // dwsc6
210             QuantizationInfo(0.036934755742549896f, 132), // dwsc7
211             QuantizationInfo(0.042609862983226776f, 94),  // dwsc8
212             QuantizationInfo(0.028358859941363335f, 127), // dwsc9
213             QuantizationInfo(0.024329448118805885f, 134), // dwsc10
214             QuantizationInfo(0.019366811960935593f, 106), // dwsc11
215             QuantizationInfo(0.007835594937205315f, 126), // dwsc12
216             QuantizationInfo(0.12616927921772003f, 211)   // dwsc13
217         };
218 
219         const std::vector<QuantizationInfo> point_weights_quant_info =
220         {
221             QuantizationInfo(0.030420949682593346f, 121), // dwsc1
222             QuantizationInfo(0.015148180536925793f, 104), // dwsc2
223             QuantizationInfo(0.013755458407104015f, 94),  // dwsc3
224             QuantizationInfo(0.007601846940815449f, 151), // dwsc4
225             QuantizationInfo(0.006431614048779011f, 122), // dwsc5
226             QuantizationInfo(0.00917122047394514f, 109),  // dwsc6
227             QuantizationInfo(0.005300046876072884f, 140), // dwsc7
228             QuantizationInfo(0.0049632852897048f, 127),   // dwsc8
229             QuantizationInfo(0.007770895957946777f, 89),  // dwsc9
230             QuantizationInfo(0.009658650495111942f, 99),  // dwsc10
231             QuantizationInfo(0.005446993745863438f, 153), // dwsc11
232             QuantizationInfo(0.00817922968417406f, 130),  // dwsc12
233             QuantizationInfo(0.018048152327537537f, 95)   // dwsc13
234         };
235 
236         graph << InputLayer(input_descriptor.set_quantization_info(in_quant_info),
237                             get_input_accessor(common_params, nullptr, false))
238               << ConvolutionLayer(
239                   3U, 3U, 32U,
240                   get_weights_accessor(data_path, "Conv2d_0_weights.npy"),
241                   get_weights_accessor(data_path, "Conv2d_0_bias.npy"),
242                   PadStrideInfo(2U, 2U, 0U, 1U, 0U, 1U, DimensionRoundingType::FLOOR),
243                   1, conv_weights_quant_info.at(0), conv_out_quant_info.at(0))
244               .set_name("Conv2d_0")
245               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 6.f)).set_name("Conv2d_0/Relu6");
246         graph << get_dwsc_node_qasymm(data_path, "Conv2d_1", 64U, PadStrideInfo(1U, 1U, 1U, 1U), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(0), point_weights_quant_info.at(0));
247         graph << get_dwsc_node_qasymm(data_path, "Conv2d_2", 128U, PadStrideInfo(2U, 2U, 0U, 1U, 0U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(1),
248                                       point_weights_quant_info.at(1));
249         graph << get_dwsc_node_qasymm(data_path, "Conv2d_3", 128U, PadStrideInfo(1U, 1U, 1U, 1U, 1U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(2),
250                                       point_weights_quant_info.at(2));
251         graph << get_dwsc_node_qasymm(data_path, "Conv2d_4", 256U, PadStrideInfo(2U, 2U, 0U, 1U, 0U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(3),
252                                       point_weights_quant_info.at(3));
253         graph << get_dwsc_node_qasymm(data_path, "Conv2d_5", 256U, PadStrideInfo(1U, 1U, 1U, 1U, 1U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(4),
254                                       point_weights_quant_info.at(4));
255         graph << get_dwsc_node_qasymm(data_path, "Conv2d_6", 512U, PadStrideInfo(2U, 2U, 0U, 1U, 0U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(5),
256                                       point_weights_quant_info.at(5));
257         graph << get_dwsc_node_qasymm(data_path, "Conv2d_7", 512U, PadStrideInfo(1U, 1U, 1U, 1U, 1U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(6),
258                                       point_weights_quant_info.at(6));
259         graph << get_dwsc_node_qasymm(data_path, "Conv2d_8", 512U, PadStrideInfo(1U, 1U, 1U, 1U, 1U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(7),
260                                       point_weights_quant_info.at(7));
261         graph << get_dwsc_node_qasymm(data_path, "Conv2d_9", 512U, PadStrideInfo(1U, 1U, 1U, 1U, 1U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(8),
262                                       point_weights_quant_info.at(8));
263         graph << get_dwsc_node_qasymm(data_path, "Conv2d_10", 512U, PadStrideInfo(1U, 1U, 1U, 1U, 1U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(9),
264                                       point_weights_quant_info.at(9));
265         graph << get_dwsc_node_qasymm(data_path, "Conv2d_11", 512U, PadStrideInfo(1U, 1U, 1U, 1U, 1U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(10),
266                                       point_weights_quant_info.at(10));
267         graph << get_dwsc_node_qasymm(data_path, "Conv2d_12", 1024U, PadStrideInfo(2U, 2U, 0U, 1U, 0U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(11),
268                                       point_weights_quant_info.at(11));
269         graph << get_dwsc_node_qasymm(data_path, "Conv2d_13", 1024U, PadStrideInfo(1U, 1U, 1U, 1U, 1U, 1U, DimensionRoundingType::FLOOR), PadStrideInfo(1U, 1U, 0U, 0U), depth_weights_quant_info.at(12),
270                                       point_weights_quant_info.at(12))
271               << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, common_params.data_layout)).set_name("Logits/AvgPool_1a")
272               << ConvolutionLayer(
273                   1U, 1U, 1001U,
274                   get_weights_accessor(data_path, "Logits_Conv2d_1c_1x1_weights.npy"),
275                   get_weights_accessor(data_path, "Logits_Conv2d_1c_1x1_bias.npy"),
276                   PadStrideInfo(1U, 1U, 0U, 0U), 1, conv_weights_quant_info.at(1), conv_out_quant_info.at(1))
277               .set_name("Logits/Conv2d_1c_1x1");
278     }
279 
get_dwsc_node_float(const std::string & data_path,std::string && param_path,unsigned int conv_filt,PadStrideInfo dwc_pad_stride_info,PadStrideInfo conv_pad_stride_info)280     ConcatLayer get_dwsc_node_float(const std::string &data_path, std::string &&param_path,
281                                     unsigned int  conv_filt,
282                                     PadStrideInfo dwc_pad_stride_info, PadStrideInfo conv_pad_stride_info)
283     {
284         std::string total_path = param_path + "_";
285         SubStream   sg(graph);
286         sg << DepthwiseConvolutionLayer(
287                3U, 3U,
288                get_weights_accessor(data_path, total_path + "depthwise_depthwise_weights.npy", DataLayout::NCHW),
289                std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
290                dwc_pad_stride_info)
291            .set_name(total_path + "depthwise/depthwise")
292            << BatchNormalizationLayer(
293                get_weights_accessor(data_path, total_path + "depthwise_BatchNorm_moving_mean.npy"),
294                get_weights_accessor(data_path, total_path + "depthwise_BatchNorm_moving_variance.npy"),
295                get_weights_accessor(data_path, total_path + "depthwise_BatchNorm_gamma.npy"),
296                get_weights_accessor(data_path, total_path + "depthwise_BatchNorm_beta.npy"),
297                0.001f)
298            .set_name(total_path + "depthwise/BatchNorm")
299            << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.f)).set_name(total_path + "depthwise/Relu6")
300            << ConvolutionLayer(
301                1U, 1U, conv_filt,
302                get_weights_accessor(data_path, total_path + "pointwise_weights.npy", DataLayout::NCHW),
303                std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr),
304                conv_pad_stride_info)
305            .set_name(total_path + "pointwise/Conv2D")
306            << BatchNormalizationLayer(
307                get_weights_accessor(data_path, total_path + "pointwise_BatchNorm_moving_mean.npy"),
308                get_weights_accessor(data_path, total_path + "pointwise_BatchNorm_moving_variance.npy"),
309                get_weights_accessor(data_path, total_path + "pointwise_BatchNorm_gamma.npy"),
310                get_weights_accessor(data_path, total_path + "pointwise_BatchNorm_beta.npy"),
311                0.001f)
312            .set_name(total_path + "pointwise/BatchNorm")
313            << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.f)).set_name(total_path + "pointwise/Relu6");
314 
315         return ConcatLayer(std::move(sg));
316     }
317 
get_dwsc_node_qasymm(const std::string & data_path,std::string && param_path,const unsigned int conv_filt,PadStrideInfo dwc_pad_stride_info,PadStrideInfo conv_pad_stride_info,QuantizationInfo depth_weights_quant_info,QuantizationInfo point_weights_quant_info)318     ConcatLayer get_dwsc_node_qasymm(const std::string &data_path, std::string &&param_path,
319                                      const unsigned int conv_filt,
320                                      PadStrideInfo dwc_pad_stride_info, PadStrideInfo conv_pad_stride_info,
321                                      QuantizationInfo depth_weights_quant_info, QuantizationInfo point_weights_quant_info)
322     {
323         std::string total_path = param_path + "_";
324         SubStream   sg(graph);
325 
326         sg << DepthwiseConvolutionLayer(
327                3U, 3U,
328                get_weights_accessor(data_path, total_path + "depthwise_weights.npy"),
329                get_weights_accessor(data_path, total_path + "depthwise_bias.npy"),
330                dwc_pad_stride_info, 1, std::move(depth_weights_quant_info))
331            .set_name(total_path + "depthwise/depthwise")
332            << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 6.f)).set_name(total_path + "depthwise/Relu6")
333            << ConvolutionLayer(
334                1U, 1U, conv_filt,
335                get_weights_accessor(data_path, total_path + "pointwise_weights.npy"),
336                get_weights_accessor(data_path, total_path + "pointwise_bias.npy"),
337                conv_pad_stride_info, 1, std::move(point_weights_quant_info))
338            .set_name(total_path + "pointwise/Conv2D")
339            << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LU_BOUNDED_RELU, 6.f)).set_name(total_path + "pointwise/Relu6");
340 
341         return ConcatLayer(std::move(sg));
342     }
343 };
344 
345 /** Main program for MobileNetV1
346  *
347  * Model is based on:
348  *      https://arxiv.org/abs/1704.04861
349  *      "MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications"
350  *      Andrew G. Howard, Menglong Zhu, Bo Chen, Dmitry Kalenichenko, Weijun Wang, Tobias Weyand, Marco Andreetto, Hartwig Adam
351  *
352  * Provenance: download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_1.0_224.tgz
353  *             download.tensorflow.org/models/mobilenet_v1_2018_08_02/mobilenet_v1_0.75_160.tgz
354  *
355  * @note To list all the possible arguments execute the binary appended with the --help option
356  *
357  * @param[in] argc Number of arguments
358  * @param[in] argv Arguments
359  */
main(int argc,char ** argv)360 int main(int argc, char **argv)
361 {
362     return arm_compute::utils::run_example<GraphMobilenetExample>(argc, argv);
363 }
364