Pablo Tello | fea8ec3 | 2018-11-16 13:25:30 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2018 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 MobileNetSSD's network using the Compute Library's graph API */ |
| 36 | class GraphSSDMobilenetExample : public Example |
| 37 | { |
| 38 | public: |
| 39 | GraphSSDMobilenetExample() |
| 40 | : cmd_parser(), common_opts(cmd_parser), common_params(), graph(0, "MobileNetSSD") |
| 41 | { |
Isabella Gottardi | 7234ed8 | 2018-11-27 08:51:10 +0000 | [diff] [blame] | 42 | // Add topk option |
| 43 | keep_topk_opt = cmd_parser.add_option<SimpleOption<int>>("topk", 100); |
| 44 | keep_topk_opt->set_help("Top k detections results per image."); |
Pablo Tello | fea8ec3 | 2018-11-16 13:25:30 +0000 | [diff] [blame] | 45 | } |
| 46 | GraphSSDMobilenetExample(const GraphSSDMobilenetExample &) = delete; |
| 47 | GraphSSDMobilenetExample &operator=(const GraphSSDMobilenetExample &) = delete; |
| 48 | GraphSSDMobilenetExample(GraphSSDMobilenetExample &&) = default; // NOLINT |
| 49 | GraphSSDMobilenetExample &operator=(GraphSSDMobilenetExample &&) = default; // NOLINT |
| 50 | ~GraphSSDMobilenetExample() override = default; |
| 51 | bool do_setup(int argc, char **argv) override |
| 52 | { |
| 53 | // Parse arguments |
| 54 | cmd_parser.parse(argc, argv); |
| 55 | |
| 56 | // Consume common parameters |
| 57 | common_params = consume_common_graph_parameters(common_opts); |
| 58 | |
| 59 | // Return when help menu is requested |
| 60 | if(common_params.help) |
| 61 | { |
| 62 | cmd_parser.print_help(argv[0]); |
| 63 | return false; |
| 64 | } |
| 65 | |
| 66 | // Print parameter values |
| 67 | std::cout << common_params << std::endl; |
| 68 | |
| 69 | // Create input descriptor |
| 70 | const TensorShape tensor_shape = permute_shape(TensorShape(300, 300, 3U, 1U), DataLayout::NCHW, common_params.data_layout); |
| 71 | TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout); |
| 72 | |
| 73 | // Set graph hints |
| 74 | graph << common_params.target |
| 75 | << DepthwiseConvolutionMethod::Optimized3x3 // FIXME(COMPMID-1073): Add heuristics to automatically call the optimized 3x3 method |
| 76 | << common_params.fast_math_hint; |
| 77 | |
| 78 | // Create core graph |
| 79 | std::string model_path = "/cnn_data/ssd_mobilenet_model/"; |
| 80 | |
| 81 | // Create a preprocessor object |
| 82 | const std::array<float, 3> mean_rgb{ { 127.5f, 127.5f, 127.5f } }; |
| 83 | std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb, 0.007843f); |
| 84 | |
| 85 | // Get trainable parameters data path |
| 86 | std::string data_path = common_params.data_path; |
| 87 | |
| 88 | // Add model path to data path |
| 89 | if(!data_path.empty()) |
| 90 | { |
| 91 | data_path += model_path; |
| 92 | } |
| 93 | |
| 94 | graph << InputLayer(input_descriptor, |
| 95 | get_input_accessor(common_params, std::move(preprocessor))); |
| 96 | |
| 97 | SubStream conv_11(graph); |
| 98 | conv_11 << ConvolutionLayer( |
| 99 | 3U, 3U, 32U, |
| 100 | get_weights_accessor(data_path, "conv0_w.npy"), |
| 101 | std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| 102 | PadStrideInfo(2, 2, 1, 1)) |
| 103 | .set_name("conv0"); |
| 104 | conv_11 << BatchNormalizationLayer(get_weights_accessor(data_path, "conv0_bn_mean.npy"), |
| 105 | get_weights_accessor(data_path, "conv0_bn_var.npy"), |
| 106 | get_weights_accessor(data_path, "conv0_scale_w.npy"), |
| 107 | get_weights_accessor(data_path, "conv0_scale_b.npy"), 0.00001f) |
| 108 | .set_name("conv0/bn") |
| 109 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv0/relu"); |
| 110 | |
| 111 | conv_11 << get_node_A(conv_11, data_path, "conv1", 64, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 112 | conv_11 << get_node_A(conv_11, data_path, "conv2", 128, PadStrideInfo(2, 2, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 113 | conv_11 << get_node_A(conv_11, data_path, "conv3", 128, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 114 | conv_11 << get_node_A(conv_11, data_path, "conv4", 256, PadStrideInfo(2, 2, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 115 | conv_11 << get_node_A(conv_11, data_path, "conv5", 256, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 116 | conv_11 << get_node_A(conv_11, data_path, "conv6", 512, PadStrideInfo(2, 2, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 117 | conv_11 << get_node_A(conv_11, data_path, "conv7", 512, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 118 | conv_11 << get_node_A(conv_11, data_path, "conv8", 512, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 119 | conv_11 << get_node_A(conv_11, data_path, "conv9", 512, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 120 | conv_11 << get_node_A(conv_11, data_path, "conv10", 512, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 121 | conv_11 << get_node_A(conv_11, data_path, "conv11", 512, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 122 | |
| 123 | SubStream conv_13(conv_11); |
| 124 | conv_13 << get_node_A(conv_11, data_path, "conv12", 1024, PadStrideInfo(2, 2, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 125 | conv_13 << get_node_A(conv_13, data_path, "conv13", 1024, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 126 | |
| 127 | SubStream conv_14(conv_13); |
| 128 | conv_14 << get_node_B(conv_13, data_path, "conv14", 512, PadStrideInfo(1, 1, 0, 0), PadStrideInfo(2, 2, 1, 1)); |
| 129 | |
| 130 | SubStream conv_15(conv_14); |
| 131 | conv_15 << get_node_B(conv_14, data_path, "conv15", 256, PadStrideInfo(1, 1, 0, 0), PadStrideInfo(2, 2, 1, 1)); |
| 132 | |
| 133 | SubStream conv_16(conv_15); |
| 134 | conv_16 << get_node_B(conv_15, data_path, "conv16", 256, PadStrideInfo(1, 1, 0, 0), PadStrideInfo(2, 2, 1, 1)); |
| 135 | |
| 136 | SubStream conv_17(conv_16); |
| 137 | conv_17 << get_node_B(conv_16, data_path, "conv17", 128, PadStrideInfo(1, 1, 0, 0), PadStrideInfo(2, 2, 1, 1)); |
| 138 | |
| 139 | //mbox_loc |
| 140 | SubStream conv_11_mbox_loc(conv_11); |
| 141 | conv_11_mbox_loc << get_node_C(conv_11, data_path, "conv11_mbox_loc", 12, PadStrideInfo(1, 1, 0, 0)); |
| 142 | |
| 143 | SubStream conv_13_mbox_loc(conv_13); |
| 144 | conv_13_mbox_loc << get_node_C(conv_13, data_path, "conv13_mbox_loc", 24, PadStrideInfo(1, 1, 0, 0)); |
| 145 | |
| 146 | SubStream conv_14_2_mbox_loc(conv_14); |
| 147 | conv_14_2_mbox_loc << get_node_C(conv_14, data_path, "conv14_2_mbox_loc", 24, PadStrideInfo(1, 1, 0, 0)); |
| 148 | |
| 149 | SubStream conv_15_2_mbox_loc(conv_15); |
| 150 | conv_15_2_mbox_loc << get_node_C(conv_15, data_path, "conv15_2_mbox_loc", 24, PadStrideInfo(1, 1, 0, 0)); |
| 151 | |
| 152 | SubStream conv_16_2_mbox_loc(conv_16); |
| 153 | conv_16_2_mbox_loc << get_node_C(conv_16, data_path, "conv16_2_mbox_loc", 24, PadStrideInfo(1, 1, 0, 0)); |
| 154 | |
| 155 | SubStream conv_17_2_mbox_loc(conv_17); |
| 156 | conv_17_2_mbox_loc << get_node_C(conv_17, data_path, "conv17_2_mbox_loc", 24, PadStrideInfo(1, 1, 0, 0)); |
| 157 | |
| 158 | SubStream mbox_loc(graph); |
| 159 | mbox_loc << ConcatLayer(std::move(conv_11_mbox_loc), std::move(conv_13_mbox_loc), conv_14_2_mbox_loc, std::move(conv_15_2_mbox_loc), |
| 160 | std::move(conv_16_2_mbox_loc), std::move(conv_17_2_mbox_loc)); |
| 161 | |
Pablo Tello | fea8ec3 | 2018-11-16 13:25:30 +0000 | [diff] [blame] | 162 | //mbox_conf |
| 163 | SubStream conv_11_mbox_conf(conv_11); |
| 164 | conv_11_mbox_conf << get_node_C(conv_11, data_path, "conv11_mbox_conf", 63, PadStrideInfo(1, 1, 0, 0)); |
| 165 | |
| 166 | SubStream conv_13_mbox_conf(conv_13); |
| 167 | conv_13_mbox_conf << get_node_C(conv_13, data_path, "conv13_mbox_conf", 126, PadStrideInfo(1, 1, 0, 0)); |
| 168 | |
| 169 | SubStream conv_14_2_mbox_conf(conv_14); |
| 170 | conv_14_2_mbox_conf << get_node_C(conv_14, data_path, "conv14_2_mbox_conf", 126, PadStrideInfo(1, 1, 0, 0)); |
| 171 | |
| 172 | SubStream conv_15_2_mbox_conf(conv_15); |
| 173 | conv_15_2_mbox_conf << get_node_C(conv_15, data_path, "conv15_2_mbox_conf", 126, PadStrideInfo(1, 1, 0, 0)); |
| 174 | |
| 175 | SubStream conv_16_2_mbox_conf(conv_16); |
| 176 | conv_16_2_mbox_conf << get_node_C(conv_16, data_path, "conv16_2_mbox_conf", 126, PadStrideInfo(1, 1, 0, 0)); |
| 177 | |
| 178 | SubStream conv_17_2_mbox_conf(conv_17); |
| 179 | conv_17_2_mbox_conf << get_node_C(conv_17, data_path, "conv17_2_mbox_conf", 126, PadStrideInfo(1, 1, 0, 0)); |
| 180 | |
| 181 | SubStream mbox_conf(graph); |
| 182 | mbox_conf << ConcatLayer(std::move(conv_11_mbox_conf), std::move(conv_13_mbox_conf), std::move(conv_14_2_mbox_conf), |
| 183 | std::move(conv_15_2_mbox_conf), std::move(conv_16_2_mbox_conf), std::move(conv_17_2_mbox_conf)); |
| 184 | mbox_conf << ReshapeLayer(TensorShape(21U, 1917U)).set_name("mbox_conf/reshape"); |
| 185 | mbox_conf << SoftmaxLayer().set_name("mbox_conf/softmax"); |
| 186 | mbox_conf << FlattenLayer().set_name("mbox_conf/flat"); |
| 187 | |
Pablo Tello | fea8ec3 | 2018-11-16 13:25:30 +0000 | [diff] [blame] | 188 | const std::vector<float> priorbox_variances = { 0.1f, 0.1f, 0.2f, 0.2f }; |
| 189 | const float priorbox_offset = 0.5f; |
| 190 | const std::vector<float> priorbox_aspect_ratios = { 2.f, 3.f }; |
| 191 | |
| 192 | //mbox_priorbox branch |
| 193 | SubStream conv_11_mbox_priorbox(conv_11); |
| 194 | |
| 195 | conv_11_mbox_priorbox << PriorBoxLayer(SubStream(graph), |
| 196 | PriorBoxLayerInfo({ 60.f }, priorbox_variances, priorbox_offset, true, false, {}, { 2.f })) |
| 197 | .set_name("conv11/priorbox"); |
| 198 | |
| 199 | SubStream conv_13_mbox_priorbox(conv_13); |
| 200 | conv_13_mbox_priorbox << PriorBoxLayer(SubStream(graph), |
| 201 | PriorBoxLayerInfo({ 105.f }, priorbox_variances, priorbox_offset, true, false, { 150.f }, priorbox_aspect_ratios)) |
| 202 | .set_name("conv13/priorbox"); |
| 203 | |
| 204 | SubStream conv_14_2_mbox_priorbox(conv_14); |
| 205 | conv_14_2_mbox_priorbox << PriorBoxLayer(SubStream(graph), |
| 206 | PriorBoxLayerInfo({ 150.f }, priorbox_variances, priorbox_offset, true, false, { 195.f }, priorbox_aspect_ratios)) |
| 207 | .set_name("conv14/priorbox"); |
| 208 | |
| 209 | SubStream conv_15_2_mbox_priorbox(conv_15); |
| 210 | conv_15_2_mbox_priorbox << PriorBoxLayer(SubStream(graph), |
| 211 | PriorBoxLayerInfo({ 195.f }, priorbox_variances, priorbox_offset, true, false, { 240.f }, priorbox_aspect_ratios)) |
| 212 | .set_name("conv15/priorbox"); |
| 213 | |
| 214 | SubStream conv_16_2_mbox_priorbox(conv_16); |
| 215 | conv_16_2_mbox_priorbox << PriorBoxLayer(SubStream(graph), |
| 216 | PriorBoxLayerInfo({ 240.f }, priorbox_variances, priorbox_offset, true, false, { 285.f }, priorbox_aspect_ratios)) |
| 217 | .set_name("conv16/priorbox"); |
| 218 | |
| 219 | SubStream conv_17_2_mbox_priorbox(conv_17); |
| 220 | conv_17_2_mbox_priorbox << PriorBoxLayer(SubStream(graph), |
| 221 | PriorBoxLayerInfo({ 285.f }, priorbox_variances, priorbox_offset, true, false, { 300.f }, priorbox_aspect_ratios)) |
| 222 | .set_name("conv17/priorbox"); |
| 223 | |
| 224 | SubStream mbox_priorbox(graph); |
| 225 | |
| 226 | mbox_priorbox << ConcatLayer( |
| 227 | (common_params.data_layout == DataLayout::NCHW) ? DataLayoutDimension::WIDTH : DataLayoutDimension::CHANNEL, |
| 228 | std::move(conv_11_mbox_priorbox), std::move(conv_13_mbox_priorbox), std::move(conv_14_2_mbox_priorbox), |
| 229 | std::move(conv_15_2_mbox_priorbox), std::move(conv_16_2_mbox_priorbox), std::move(conv_17_2_mbox_priorbox)); |
| 230 | |
Isabella Gottardi | 7234ed8 | 2018-11-27 08:51:10 +0000 | [diff] [blame] | 231 | const int num_classes = 21; |
| 232 | const bool share_location = true; |
| 233 | const DetectionOutputLayerCodeType detection_type = DetectionOutputLayerCodeType::CENTER_SIZE; |
| 234 | const int keep_top_k = keep_topk_opt->value(); |
| 235 | const float nms_threshold = 0.45f; |
| 236 | const int label_id_background = 0; |
| 237 | const float conf_thrs = 0.25f; |
| 238 | const int top_k = 100; |
| 239 | |
| 240 | SubStream detection_ouput(mbox_loc); |
| 241 | detection_ouput << DetectionOutputLayer(std::move(mbox_conf), std::move(mbox_priorbox), |
| 242 | DetectionOutputLayerInfo(num_classes, share_location, detection_type, keep_top_k, nms_threshold, top_k, label_id_background, conf_thrs)); |
| 243 | detection_ouput << OutputLayer(get_detection_output_accessor(common_params, { tensor_shape })); |
Pablo Tello | fea8ec3 | 2018-11-16 13:25:30 +0000 | [diff] [blame] | 244 | |
| 245 | // Finalize graph |
| 246 | GraphConfig config; |
| 247 | config.num_threads = common_params.threads; |
| 248 | config.use_tuner = common_params.enable_tuner; |
| 249 | config.tuner_file = common_params.tuner_file; |
| 250 | |
| 251 | graph.finalize(common_params.target, config); |
| 252 | |
| 253 | return true; |
| 254 | } |
| 255 | void do_run() override |
| 256 | { |
| 257 | // Run graph |
| 258 | graph.run(); |
| 259 | } |
| 260 | |
| 261 | private: |
| 262 | CommandLineParser cmd_parser; |
| 263 | CommonGraphOptions common_opts; |
Isabella Gottardi | 7234ed8 | 2018-11-27 08:51:10 +0000 | [diff] [blame] | 264 | SimpleOption<int> *keep_topk_opt{ nullptr }; |
| 265 | CommonGraphParams common_params; |
| 266 | Stream graph; |
Pablo Tello | fea8ec3 | 2018-11-16 13:25:30 +0000 | [diff] [blame] | 267 | |
| 268 | ConcatLayer get_node_A(IStream &master_graph, const std::string &data_path, std::string &¶m_path, |
| 269 | unsigned int conv_filt, |
| 270 | PadStrideInfo dwc_pad_stride_info, PadStrideInfo conv_pad_stride_info) |
| 271 | { |
| 272 | const std::string total_path = param_path + "_"; |
| 273 | SubStream sg(master_graph); |
| 274 | |
| 275 | sg << DepthwiseConvolutionLayer( |
| 276 | 3U, 3U, |
| 277 | get_weights_accessor(data_path, total_path + "dw_w.npy"), |
| 278 | std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| 279 | dwc_pad_stride_info) |
| 280 | .set_name(param_path + "/dw") |
| 281 | << BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "dw_bn_mean.npy"), |
| 282 | get_weights_accessor(data_path, total_path + "dw_bn_var.npy"), |
| 283 | get_weights_accessor(data_path, total_path + "dw_scale_w.npy"), |
| 284 | get_weights_accessor(data_path, total_path + "dw_scale_b.npy"), 0.00001f) |
| 285 | .set_name(param_path + "/dw/bn") |
| 286 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "dw/relu") |
| 287 | |
| 288 | << ConvolutionLayer( |
| 289 | 1U, 1U, conv_filt, |
| 290 | get_weights_accessor(data_path, total_path + "w.npy"), |
| 291 | std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| 292 | conv_pad_stride_info) |
| 293 | .set_name(param_path + "/pw") |
| 294 | << BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "bn_mean.npy"), |
| 295 | get_weights_accessor(data_path, total_path + "bn_var.npy"), |
| 296 | get_weights_accessor(data_path, total_path + "scale_w.npy"), |
| 297 | get_weights_accessor(data_path, total_path + "scale_b.npy"), 0.00001f) |
| 298 | .set_name(param_path + "/pw/bn") |
| 299 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "pw/relu"); |
| 300 | |
| 301 | return ConcatLayer(std::move(sg)); |
| 302 | } |
| 303 | |
| 304 | ConcatLayer get_node_B(IStream &master_graph, const std::string &data_path, std::string &¶m_path, |
| 305 | unsigned int conv_filt, |
| 306 | PadStrideInfo conv_pad_stride_info_1, PadStrideInfo conv_pad_stride_info_2) |
| 307 | { |
| 308 | const std::string total_path = param_path + "_"; |
| 309 | SubStream sg(master_graph); |
| 310 | |
| 311 | sg << ConvolutionLayer( |
| 312 | 1, 1, conv_filt / 2, |
| 313 | get_weights_accessor(data_path, total_path + "1_w.npy"), |
| 314 | std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| 315 | conv_pad_stride_info_1) |
| 316 | .set_name(total_path + "1/conv") |
| 317 | << BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "1_bn_mean.npy"), |
| 318 | get_weights_accessor(data_path, total_path + "1_bn_var.npy"), |
| 319 | get_weights_accessor(data_path, total_path + "1_scale_w.npy"), |
| 320 | get_weights_accessor(data_path, total_path + "1_scale_b.npy"), 0.00001f) |
| 321 | .set_name(total_path + "1/bn") |
| 322 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(total_path + "1/relu"); |
| 323 | |
| 324 | sg << ConvolutionLayer( |
| 325 | 3, 3, conv_filt, |
| 326 | get_weights_accessor(data_path, total_path + "2_w.npy"), |
| 327 | std::unique_ptr<arm_compute::graph::ITensorAccessor>(nullptr), |
| 328 | conv_pad_stride_info_2) |
| 329 | .set_name(total_path + "2/conv") |
| 330 | << BatchNormalizationLayer(get_weights_accessor(data_path, total_path + "2_bn_mean.npy"), |
| 331 | get_weights_accessor(data_path, total_path + "2_bn_var.npy"), |
| 332 | get_weights_accessor(data_path, total_path + "2_scale_w.npy"), |
| 333 | get_weights_accessor(data_path, total_path + "2_scale_b.npy"), 0.00001f) |
| 334 | .set_name(total_path + "2/bn") |
| 335 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(total_path + "2/relu"); |
| 336 | |
| 337 | return ConcatLayer(std::move(sg)); |
| 338 | } |
| 339 | |
| 340 | ConcatLayer get_node_C(IStream &master_graph, const std::string &data_path, std::string &¶m_path, |
| 341 | unsigned int conv_filt, PadStrideInfo conv_pad_stride_info) |
| 342 | { |
| 343 | const std::string total_path = param_path + "_"; |
| 344 | SubStream sg(master_graph); |
| 345 | sg << ConvolutionLayer( |
| 346 | 1U, 1U, conv_filt, |
| 347 | get_weights_accessor(data_path, total_path + "w.npy"), |
| 348 | get_weights_accessor(data_path, total_path + "b.npy"), |
| 349 | conv_pad_stride_info) |
| 350 | .set_name(param_path + "/conv"); |
| 351 | if(common_params.data_layout == DataLayout::NCHW) |
| 352 | { |
| 353 | sg << PermuteLayer(PermutationVector(2U, 0U, 1U), DataLayout::NHWC).set_name(param_path + "/perm"); |
| 354 | } |
| 355 | sg << FlattenLayer().set_name(param_path + "/flat"); |
| 356 | |
| 357 | return ConcatLayer(std::move(sg)); |
| 358 | } |
| 359 | }; |
| 360 | |
| 361 | /** Main program for MobileNetSSD |
| 362 | * |
| 363 | * Model is based on: |
| 364 | * http://arxiv.org/abs/1512.02325 |
| 365 | * SSD: Single Shot MultiBox Detector |
| 366 | * Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed, Cheng-Yang Fu, Alexander C. Berg |
| 367 | * |
| 368 | * @note To list all the possible arguments execute the binary appended with the --help option |
| 369 | * |
| 370 | * @param[in] argc Number of arguments |
| 371 | * @param[in] argv Arguments |
| 372 | */ |
| 373 | int main(int argc, char **argv) |
| 374 | { |
| 375 | return arm_compute::utils::run_example<GraphSSDMobilenetExample>(argc, argv); |
| 376 | } |