Isabella Gottardi | bc4484a | 2018-02-02 11:27:32 +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/Graph.h" |
| 25 | #include "arm_compute/graph/Nodes.h" |
| 26 | #include "arm_compute/graph/SubGraph.h" |
| 27 | #include "support/ToolchainSupport.h" |
| 28 | #include "utils/GraphUtils.h" |
| 29 | #include "utils/Utils.h" |
| 30 | |
| 31 | #include <cstdlib> |
| 32 | #include <tuple> |
| 33 | |
| 34 | using namespace arm_compute::utils; |
| 35 | using namespace arm_compute::graph; |
| 36 | using namespace arm_compute::graph_utils; |
| 37 | using namespace arm_compute::logging; |
| 38 | |
| 39 | namespace |
| 40 | { |
| 41 | } // namespace |
| 42 | |
| 43 | /** Example demonstrating how to implement Squeezenet's v1.1 network using the Compute Library's graph API |
| 44 | * |
| 45 | * @param[in] argc Number of arguments |
| 46 | * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL), [optional] Path to the weights folder, [optional] image, [optional] labels ) |
| 47 | */ |
| 48 | class GraphSqueezenet_v1_1Example : public Example |
| 49 | { |
| 50 | public: |
| 51 | void do_setup(int argc, char **argv) override |
| 52 | { |
| 53 | std::string data_path; /* Path to the trainable data */ |
| 54 | std::string image; /* Image data */ |
| 55 | std::string label; /* Label data */ |
| 56 | |
Georgios Pinitas | 140fdc7 | 2018-02-16 11:42:38 +0000 | [diff] [blame] | 57 | // Create a preprocessor object |
| 58 | const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } }; |
| 59 | std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb); |
Isabella Gottardi | bc4484a | 2018-02-02 11:27:32 +0000 | [diff] [blame] | 60 | |
Michele Di Giorgio | e3fba0a | 2018-02-14 14:18:01 +0000 | [diff] [blame] | 61 | // Set target. 0 (NEON), 1 (OpenCL), 2 (OpenCL with Tuner). By default it is NEON |
| 62 | const int int_target_hint = argc > 1 ? std::strtol(argv[1], nullptr, 10) : 0; |
| 63 | TargetHint target_hint = set_target_hint(int_target_hint); |
Isabella Gottardi | bc4484a | 2018-02-02 11:27:32 +0000 | [diff] [blame] | 64 | |
| 65 | // Parse arguments |
| 66 | if(argc < 2) |
| 67 | { |
| 68 | // Print help |
| 69 | std::cout << "Usage: " << argv[0] << " [target] [path_to_data] [image] [labels]\n\n"; |
| 70 | std::cout << "No data folder provided: using random values\n\n"; |
| 71 | } |
| 72 | else if(argc == 2) |
| 73 | { |
| 74 | std::cout << "Usage: " << argv[0] << " " << argv[1] << " [path_to_data] [image] [labels]\n\n"; |
| 75 | std::cout << "No data folder provided: using random values\n\n"; |
| 76 | } |
| 77 | else if(argc == 3) |
| 78 | { |
| 79 | data_path = argv[2]; |
| 80 | std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " [image] [labels]\n\n"; |
| 81 | std::cout << "No image provided: using random values\n\n"; |
| 82 | } |
| 83 | else if(argc == 4) |
| 84 | { |
| 85 | data_path = argv[2]; |
| 86 | image = argv[3]; |
| 87 | std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " [labels]\n\n"; |
| 88 | std::cout << "No text file with labels provided: skipping output accessor\n\n"; |
| 89 | } |
| 90 | else |
| 91 | { |
| 92 | data_path = argv[2]; |
| 93 | image = argv[3]; |
| 94 | label = argv[4]; |
| 95 | } |
| 96 | |
Michele Di Giorgio | e3fba0a | 2018-02-14 14:18:01 +0000 | [diff] [blame] | 97 | // Initialize graph |
| 98 | graph.graph_init(int_target_hint == 2); |
| 99 | |
Isabella Gottardi | bc4484a | 2018-02-02 11:27:32 +0000 | [diff] [blame] | 100 | graph << target_hint |
| 101 | << Tensor(TensorInfo(TensorShape(227U, 227U, 3U, 1U), 1, DataType::F32), |
Georgios Pinitas | 140fdc7 | 2018-02-16 11:42:38 +0000 | [diff] [blame] | 102 | get_input_accessor(image, std::move(preprocessor))) |
Isabella Gottardi | bc4484a | 2018-02-02 11:27:32 +0000 | [diff] [blame] | 103 | << ConvolutionLayer( |
| 104 | 3U, 3U, 64U, |
| 105 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv1_w.npy"), |
| 106 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv1_b.npy"), |
| 107 | PadStrideInfo(2, 2, 0, 0)) |
| 108 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 109 | << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))) |
| 110 | << ConvolutionLayer( |
| 111 | 1U, 1U, 16U, |
| 112 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire2_squeeze1x1_w.npy"), |
| 113 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire2_squeeze1x1_b.npy"), |
| 114 | PadStrideInfo(1, 1, 0, 0)) |
| 115 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 116 | << get_expand_fire_node(data_path, "fire2", 64U, 64U) |
| 117 | << ConvolutionLayer( |
| 118 | 1U, 1U, 16U, |
| 119 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire3_squeeze1x1_w.npy"), |
| 120 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire3_squeeze1x1_b.npy"), |
| 121 | PadStrideInfo(1, 1, 0, 0)) |
| 122 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 123 | << get_expand_fire_node(data_path, "fire3", 64U, 64U) |
| 124 | << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))) |
| 125 | << ConvolutionLayer( |
| 126 | 1U, 1U, 32U, |
| 127 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire4_squeeze1x1_w.npy"), |
| 128 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire4_squeeze1x1_b.npy"), |
| 129 | PadStrideInfo(1, 1, 0, 0)) |
| 130 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 131 | << get_expand_fire_node(data_path, "fire4", 128U, 128U) |
| 132 | << ConvolutionLayer( |
| 133 | 1U, 1U, 32U, |
| 134 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire5_squeeze1x1_w.npy"), |
| 135 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire5_squeeze1x1_b.npy"), |
| 136 | PadStrideInfo(1, 1, 0, 0)) |
| 137 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 138 | << get_expand_fire_node(data_path, "fire5", 128U, 128U) |
| 139 | << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))) |
| 140 | << ConvolutionLayer( |
| 141 | 1U, 1U, 48U, |
| 142 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire6_squeeze1x1_w.npy"), |
| 143 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire6_squeeze1x1_b.npy"), |
| 144 | PadStrideInfo(1, 1, 0, 0)) |
| 145 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 146 | << get_expand_fire_node(data_path, "fire6", 192U, 192U) |
| 147 | << ConvolutionLayer( |
| 148 | 1U, 1U, 48U, |
| 149 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire7_squeeze1x1_w.npy"), |
| 150 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire7_squeeze1x1_b.npy"), |
| 151 | PadStrideInfo(1, 1, 0, 0)) |
| 152 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 153 | << get_expand_fire_node(data_path, "fire7", 192U, 192U) |
| 154 | << ConvolutionLayer( |
| 155 | 1U, 1U, 64U, |
| 156 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire8_squeeze1x1_w.npy"), |
| 157 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire8_squeeze1x1_b.npy"), |
| 158 | PadStrideInfo(1, 1, 0, 0)) |
| 159 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 160 | << get_expand_fire_node(data_path, "fire8", 256U, 256U) |
| 161 | << ConvolutionLayer( |
| 162 | 1U, 1U, 64U, |
| 163 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire9_squeeze1x1_w.npy"), |
| 164 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/fire9_squeeze1x1_b.npy"), |
| 165 | PadStrideInfo(1, 1, 0, 0)) |
| 166 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 167 | << get_expand_fire_node(data_path, "fire9", 256U, 256U) |
| 168 | << ConvolutionLayer( |
| 169 | 1U, 1U, 1000U, |
| 170 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv10_w.npy"), |
| 171 | get_weights_accessor(data_path, "/cnn_data/squeezenet_v1_1_model/conv10_b.npy"), |
| 172 | PadStrideInfo(1, 1, 0, 0)) |
| 173 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)) |
| 174 | << PoolingLayer(PoolingLayerInfo(PoolingType::AVG)) |
| 175 | << FlattenLayer() |
| 176 | << SoftmaxLayer() |
| 177 | << Tensor(get_output_accessor(label, 5)); |
| 178 | } |
| 179 | void do_run() override |
| 180 | { |
| 181 | // Run graph |
| 182 | graph.run(); |
| 183 | } |
| 184 | |
| 185 | private: |
| 186 | Graph graph{}; |
| 187 | |
| 188 | BranchLayer get_expand_fire_node(const std::string &data_path, std::string &¶m_path, unsigned int expand1_filt, unsigned int expand3_filt) |
| 189 | { |
| 190 | std::string total_path = "/cnn_data/squeezenet_v1_1_model/" + param_path + "_"; |
| 191 | SubGraph i_a; |
| 192 | i_a << ConvolutionLayer( |
| 193 | 1U, 1U, expand1_filt, |
| 194 | get_weights_accessor(data_path, total_path + "expand1x1_w.npy"), |
| 195 | get_weights_accessor(data_path, total_path + "expand1x1_b.npy"), |
| 196 | PadStrideInfo(1, 1, 0, 0)) |
| 197 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)); |
| 198 | |
| 199 | SubGraph i_b; |
| 200 | i_b << ConvolutionLayer( |
| 201 | 3U, 3U, expand3_filt, |
| 202 | get_weights_accessor(data_path, total_path + "expand3x3_w.npy"), |
| 203 | get_weights_accessor(data_path, total_path + "expand3x3_b.npy"), |
| 204 | PadStrideInfo(1, 1, 1, 1)) |
| 205 | << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)); |
| 206 | |
| 207 | return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b)); |
| 208 | } |
| 209 | }; |
| 210 | |
| 211 | /** Main program for Squeezenet v1.1 |
| 212 | * |
| 213 | * @param[in] argc Number of arguments |
| 214 | * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL), [optional] Path to the weights folder, [optional] image, [optional] labels ) |
| 215 | */ |
| 216 | int main(int argc, char **argv) |
| 217 | { |
| 218 | return arm_compute::utils::run_example<GraphSqueezenet_v1_1Example>(argc, argv); |
| 219 | } |