Georgios Pinitas | 93dcd83 | 2017-10-27 12:48:20 +0100 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2017 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 | #ifndef __ARM_COMPUTE_TEST_MODEL_OBJECTS_MOBILENET_H__ |
| 25 | #define __ARM_COMPUTE_TEST_MODEL_OBJECTS_MOBILENET_H__ |
| 26 | |
| 27 | #include "tests/AssetsLibrary.h" |
| 28 | #include "tests/Globals.h" |
| 29 | #include "tests/Utils.h" |
| 30 | |
| 31 | #include "utils/Utils.h" |
| 32 | |
| 33 | #include <memory> |
| 34 | |
| 35 | using namespace arm_compute; |
| 36 | using namespace arm_compute::test; |
| 37 | |
| 38 | namespace arm_compute |
| 39 | { |
| 40 | namespace test |
| 41 | { |
| 42 | namespace networks |
| 43 | { |
| 44 | /** MobileNet model object */ |
| 45 | template <typename TensorType, |
| 46 | typename Accessor, |
| 47 | typename ActivationLayerFunction, |
| 48 | typename ConvolutionLayerFunction, |
| 49 | typename DirectConvolutionLayerFunction, |
| 50 | typename DepthwiseConvolutionFunction, |
| 51 | typename ReshapeFunction, |
| 52 | typename PoolingLayerFunction> |
| 53 | class MobileNetNetwork |
| 54 | { |
| 55 | public: |
| 56 | void init(int batches) |
| 57 | { |
| 58 | _batches = batches; |
| 59 | |
| 60 | // Initialize input, output |
| 61 | input.allocator()->init(TensorInfo(TensorShape(224U, 224U, 3U, _batches), 1, DataType::F32)); |
| 62 | output.allocator()->init(TensorInfo(TensorShape(11U, _batches), 1, DataType::F32)); |
| 63 | // Initialize weights and biases |
| 64 | w_conv3x3.allocator()->init(TensorInfo(TensorShape(3U, 3U, 3U, 16U), 1, DataType::F32)); |
| 65 | b_conv3x3.allocator()->init(TensorInfo(TensorShape(16U), 1, DataType::F32)); |
| 66 | depthwise_conv_block_init(0, 16, 16); |
| 67 | depthwise_conv_block_init(1, 16, 32); |
| 68 | depthwise_conv_block_init(2, 32, 32); |
| 69 | depthwise_conv_block_init(3, 32, 64); |
| 70 | depthwise_conv_block_init(4, 64, 64); |
| 71 | depthwise_conv_block_init(5, 64, 128); |
| 72 | depthwise_conv_block_init(6, 128, 128); |
| 73 | depthwise_conv_block_init(7, 128, 128); |
| 74 | depthwise_conv_block_init(8, 128, 128); |
| 75 | depthwise_conv_block_init(9, 128, 128); |
| 76 | depthwise_conv_block_init(10, 128, 128); |
| 77 | depthwise_conv_block_init(11, 128, 256); |
| 78 | depthwise_conv_block_init(12, 256, 256); |
| 79 | w_conv[13].allocator()->init(TensorInfo(TensorShape(1U, 1U, 256U, 11U), 1, DataType::F32)); |
| 80 | b_conv[13].allocator()->init(TensorInfo(TensorShape(11U), 1, DataType::F32)); |
| 81 | } |
| 82 | |
| 83 | /** Build the model. */ |
| 84 | void build() |
| 85 | { |
| 86 | // Configure Layers |
| 87 | conv3x3.configure(&input, &w_conv3x3, &b_conv3x3, &conv_out[0], PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::FLOOR)); |
| 88 | conv3x3_act.configure(&conv_out[0], nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.f)); |
| 89 | depthwise_conv_block_build(0, PadStrideInfo(1, 1, 1, 1), PadStrideInfo(1, 1, 0, 0)); |
| 90 | depthwise_conv_block_build(1, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 91 | depthwise_conv_block_build(2, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 92 | depthwise_conv_block_build(3, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 93 | depthwise_conv_block_build(4, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 94 | depthwise_conv_block_build(5, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 95 | depthwise_conv_block_build(6, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 96 | depthwise_conv_block_build(7, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 97 | depthwise_conv_block_build(8, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 98 | depthwise_conv_block_build(9, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 99 | depthwise_conv_block_build(10, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 100 | depthwise_conv_block_build(11, PadStrideInfo(2, 2, 0, 1, 0, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 101 | depthwise_conv_block_build(12, PadStrideInfo(1, 1, 1, 1, 1, 1, DimensionRoundingType::FLOOR), PadStrideInfo(1, 1, 0, 0)); |
| 102 | pool.configure(&conv_out[13], &pool_out, PoolingLayerInfo(PoolingType::AVG, 7, PadStrideInfo(2, 2, 0, 0))); |
| 103 | conv1x1[13].configure(&pool_out, &w_conv[13], &b_conv[13], &conv_out[14], PadStrideInfo(1, 1, 0, 0)); |
| 104 | logistic.configure(&conv_out[14], nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LOGISTIC)); |
| 105 | reshape.configure(&conv_out[14], &output); |
| 106 | } |
| 107 | |
| 108 | void allocate() |
| 109 | { |
| 110 | input.allocator()->allocate(); |
| 111 | output.allocator()->allocate(); |
| 112 | |
| 113 | w_conv3x3.allocator()->allocate(); |
| 114 | b_conv3x3.allocator()->allocate(); |
| 115 | for(unsigned int i = 0; i < w_conv.size(); ++i) |
| 116 | { |
| 117 | w_conv[i].allocator()->allocate(); |
| 118 | b_conv[i].allocator()->allocate(); |
| 119 | } |
| 120 | for(unsigned int i = 0; i < w_dwc.size(); ++i) |
| 121 | { |
| 122 | w_dwc[i].allocator()->allocate(); |
| 123 | b_dwc[i].allocator()->allocate(); |
| 124 | } |
| 125 | for(auto &o : conv_out) |
| 126 | { |
| 127 | o.allocator()->allocate(); |
| 128 | } |
| 129 | for(auto &o : dwc_out) |
| 130 | { |
| 131 | o.allocator()->allocate(); |
| 132 | } |
| 133 | pool_out.allocator()->allocate(); |
| 134 | } |
| 135 | |
| 136 | /** Fills the trainable parameters and input with random data. */ |
| 137 | void fill_random() |
| 138 | { |
| 139 | unsigned int seed_idx = 0; |
| 140 | std::uniform_real_distribution<> distribution(-1, 1); |
| 141 | library->fill(Accessor(input), distribution, seed_idx++); |
| 142 | |
| 143 | library->fill(Accessor(w_conv3x3), distribution, seed_idx++); |
| 144 | library->fill(Accessor(b_conv3x3), distribution, seed_idx++); |
| 145 | for(unsigned int i = 0; i < w_conv.size(); ++i) |
| 146 | { |
| 147 | library->fill(Accessor(w_conv[i]), distribution, seed_idx++); |
| 148 | library->fill(Accessor(b_conv[i]), distribution, seed_idx++); |
| 149 | } |
| 150 | for(unsigned int i = 0; i < w_dwc.size(); ++i) |
| 151 | { |
| 152 | library->fill(Accessor(w_dwc[i]), distribution, seed_idx++); |
| 153 | library->fill(Accessor(b_dwc[i]), distribution, seed_idx++); |
| 154 | } |
| 155 | } |
| 156 | |
| 157 | /** Feed input to network from file. |
| 158 | * |
| 159 | * @param name File name of containing the input data. |
| 160 | */ |
| 161 | void feed(std::string name) |
| 162 | { |
| 163 | library->fill_layer_data(Accessor(input), name); |
| 164 | } |
| 165 | |
| 166 | /** Get the classification results. |
| 167 | * |
| 168 | * @return Vector containing the classified labels |
| 169 | */ |
| 170 | std::vector<unsigned int> get_classifications() |
| 171 | { |
| 172 | std::vector<unsigned int> classified_labels; |
| 173 | Accessor output_accessor(output); |
| 174 | |
| 175 | Window window; |
| 176 | window.set(Window::DimX, Window::Dimension(0, 1, 1)); |
| 177 | for(unsigned int d = 1; d < output_accessor.shape().num_dimensions(); ++d) |
| 178 | { |
| 179 | window.set(d, Window::Dimension(0, output_accessor.shape()[d], 1)); |
| 180 | } |
| 181 | |
| 182 | execute_window_loop(window, [&](const Coordinates & id) |
| 183 | { |
| 184 | int max_idx = 0; |
| 185 | float val = 0; |
| 186 | const void *const out_ptr = output_accessor(id); |
| 187 | for(unsigned int l = 0; l < output_accessor.shape().x(); ++l) |
| 188 | { |
| 189 | float curr_val = reinterpret_cast<const float *>(out_ptr)[l]; |
| 190 | if(curr_val > val) |
| 191 | { |
| 192 | max_idx = l; |
| 193 | val = curr_val; |
| 194 | } |
| 195 | } |
| 196 | classified_labels.push_back(max_idx); |
| 197 | }); |
| 198 | return classified_labels; |
| 199 | } |
| 200 | |
| 201 | /** Clear all allocated memory from the tensor objects */ |
| 202 | void clear() |
| 203 | { |
| 204 | input.allocator()->free(); |
| 205 | output.allocator()->free(); |
| 206 | |
| 207 | w_conv3x3.allocator()->free(); |
| 208 | b_conv3x3.allocator()->free(); |
| 209 | for(unsigned int i = 0; i < w_conv.size(); ++i) |
| 210 | { |
| 211 | w_conv[i].allocator()->free(); |
| 212 | b_conv[i].allocator()->free(); |
| 213 | } |
| 214 | for(unsigned int i = 0; i < w_dwc.size(); ++i) |
| 215 | { |
| 216 | w_dwc[i].allocator()->free(); |
| 217 | b_dwc[i].allocator()->free(); |
| 218 | } |
| 219 | for(auto &o : conv_out) |
| 220 | { |
| 221 | o.allocator()->free(); |
| 222 | } |
| 223 | for(auto &o : dwc_out) |
| 224 | { |
| 225 | o.allocator()->free(); |
| 226 | } |
| 227 | pool_out.allocator()->free(); |
| 228 | } |
| 229 | |
| 230 | /** Runs the model */ |
| 231 | void run() |
| 232 | { |
| 233 | conv3x3.run(); |
| 234 | conv3x3_act.run(); |
| 235 | depthwise_conv_block_run(0); |
| 236 | depthwise_conv_block_run(1); |
| 237 | depthwise_conv_block_run(2); |
| 238 | depthwise_conv_block_run(3); |
| 239 | depthwise_conv_block_run(4); |
| 240 | depthwise_conv_block_run(5); |
| 241 | depthwise_conv_block_run(6); |
| 242 | depthwise_conv_block_run(7); |
| 243 | depthwise_conv_block_run(8); |
| 244 | depthwise_conv_block_run(9); |
| 245 | depthwise_conv_block_run(10); |
| 246 | depthwise_conv_block_run(11); |
| 247 | depthwise_conv_block_run(12); |
| 248 | pool.run(); |
| 249 | conv1x1[13].run(); |
| 250 | logistic.run(); |
| 251 | reshape.run(); |
| 252 | } |
| 253 | |
| 254 | private: |
| 255 | void depthwise_conv_block_init(unsigned int idx, unsigned int ifm, unsigned int ofm) |
| 256 | { |
| 257 | w_dwc[idx].allocator()->init(TensorInfo(TensorShape(3U, 3U, ifm), 1, DataType::F32)); |
| 258 | b_dwc[idx].allocator()->init(TensorInfo(TensorShape(ifm), 1, DataType::F32)); |
| 259 | w_conv[idx].allocator()->init(TensorInfo(TensorShape(1U, 1U, ifm, ofm), 1, DataType::F32)); |
| 260 | b_conv[idx].allocator()->init(TensorInfo(TensorShape(ofm), 1, DataType::F32)); |
| 261 | } |
| 262 | void depthwise_conv_block_build(unsigned int idx, PadStrideInfo dwc_ps, PadStrideInfo conv_ps) |
| 263 | { |
| 264 | dwc3x3[idx].configure(&conv_out[idx], &w_dwc[idx], &b_dwc[idx], &dwc_out[idx], dwc_ps); |
| 265 | act[2 * idx].configure(&dwc_out[idx], nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.f)); |
| 266 | conv1x1[idx].configure(&dwc_out[idx], &w_conv[idx], &b_conv[idx], &conv_out[idx + 1], conv_ps); |
| 267 | act[2 * idx + 1].configure(&conv_out[idx], nullptr, ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::BOUNDED_RELU, 6.f)); |
| 268 | } |
| 269 | void depthwise_conv_block_run(unsigned int idx) |
| 270 | { |
| 271 | dwc3x3[idx].run(); |
| 272 | act[2 * idx].run(); |
| 273 | conv1x1[idx].run(); |
| 274 | act[2 * idx + 1].run(); |
| 275 | } |
| 276 | |
| 277 | private: |
| 278 | unsigned int _batches{ 0 }; |
| 279 | |
| 280 | ConvolutionLayerFunction conv3x3{}; |
| 281 | ActivationLayerFunction conv3x3_act{}; |
| 282 | std::array<ActivationLayerFunction, 26> act{ {} }; |
| 283 | std::array<DirectConvolutionLayerFunction, 14> conv1x1{ {} }; |
| 284 | std::array<DepthwiseConvolutionFunction, 13> dwc3x3{ {} }; |
| 285 | PoolingLayerFunction pool{}; |
| 286 | ActivationLayerFunction logistic{}; |
| 287 | ReshapeFunction reshape{}; |
| 288 | |
| 289 | TensorType w_conv3x3{}, b_conv3x3{}; |
| 290 | std::array<TensorType, 14> w_conv{ {} }, b_conv{ {} }; |
| 291 | std::array<TensorType, 13> w_dwc{ {} }, b_dwc{ {} }; |
| 292 | |
| 293 | TensorType input{}, output{}; |
| 294 | |
| 295 | std::array<TensorType, 15> conv_out{ {} }; |
| 296 | std::array<TensorType, 13> dwc_out{ {} }; |
| 297 | TensorType pool_out{}; |
| 298 | }; |
| 299 | } // namespace networks |
| 300 | } // namespace test |
| 301 | } // namespace arm_compute |
| 302 | #endif //__ARM_COMPUTE_TEST_MODEL_OBJECTS_MOBILENET_H__ |