examples/graph_vgg19.cpp

   1 /*
   2  * Copyright (c) 2017, 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 "support/ToolchainSupport.h"
  27 #include "utils/GraphUtils.h"
  28 #include "utils/Utils.h"
  29
  30 #include <cstdlib>
  31
  32 using namespace arm_compute::utils;
  33 using namespace arm_compute::graph;
  34 using namespace arm_compute::graph_utils;
  35
  36 /** Example demonstrating how to implement VGG19's network using the Compute Library's graph API
  37  *
  38  * @param[in] argc Number of arguments
  39  * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL), [optional] Path to the weights folder, [optional] image, [optional] labels )
  40  */
  41 class GraphVGG19Example : public Example
  42 {
  43 public:
  44     void do_setup(int argc, char **argv) override
  45     {
  46         std::string data_path; /* Path to the trainable data */
  47         std::string image;     /* Image data */
  48         std::string label;     /* Label data */
  49
  50         constexpr float mean_r = 123.68f;  /* Mean value to subtract from red channel */
  51         constexpr float mean_g = 116.779f; /* Mean value to subtract from green channel */
  52         constexpr float mean_b = 103.939f; /* Mean value to subtract from blue channel */
  53
  54         // Set target. 0 (NEON), 1 (OpenCL). By default it is NEON
  55         TargetHint            target_hint      = set_target_hint(argc > 1 ? std::strtol(argv[1], nullptr, 10) : 0);
  56         ConvolutionMethodHint convolution_hint = ConvolutionMethodHint::DIRECT;
  57
  58         // Parse arguments
  59         if(argc < 2)
  60         {
  61             // Print help
  62             std::cout << "Usage: " << argv[0] << " [target] [path_to_data] [image] [labels]\n\n";
  63             std::cout << "No data folder provided: using random values\n\n";
  64         }
  65         else if(argc == 2)
  66         {
  67             std::cout << "Usage: " << argv[0] << " " << argv[1] << " [path_to_data] [image] [labels]\n\n";
  68             std::cout << "No data folder provided: using random values\n\n";
  69         }
  70         else if(argc == 3)
  71         {
  72             data_path = argv[2];
  73             std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " [image] [labels]\n\n";
  74             std::cout << "No image provided: using random values\n\n";
  75         }
  76         else if(argc == 4)
  77         {
  78             data_path = argv[2];
  79             image     = argv[3];
  80             std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " [labels]\n\n";
  81             std::cout << "No text file with labels provided: skipping output accessor\n\n";
  82         }
  83         else
  84         {
  85             data_path = argv[2];
  86             image     = argv[3];
  87             label     = argv[4];
  88         }
  89
  90         graph << target_hint
  91               << convolution_hint
  92               << Tensor(TensorInfo(TensorShape(224U, 224U, 3U, 1U), 1, DataType::F32),
  93                         get_input_accessor(image, mean_r, mean_g, mean_b))
  94               // Layer 1
  95               << ConvolutionLayer(
  96                   3U, 3U, 64U,
  97                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv1_1_w.npy"),
  98                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv1_1_b.npy"),
  99                   PadStrideInfo(1, 1, 1, 1))
 100               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 101               << ConvolutionLayer(
 102                   3U, 3U, 64U,
 103                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv1_2_w.npy"),
 104                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv1_2_b.npy"),
 105                   PadStrideInfo(1, 1, 1, 1))
 106               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 107               << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
 108               // Layer 2
 109               << ConvolutionLayer(
 110                   3U, 3U, 128U,
 111                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv2_1_w.npy"),
 112                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv2_1_b.npy"),
 113                   PadStrideInfo(1, 1, 1, 1))
 114               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 115               << ConvolutionLayer(
 116                   3U, 3U, 128U,
 117                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv2_2_w.npy"),
 118                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv2_2_b.npy"),
 119                   PadStrideInfo(1, 1, 1, 1))
 120               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 121               << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
 122               // Layer 3
 123               << ConvolutionLayer(
 124                   3U, 3U, 256U,
 125                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_1_w.npy"),
 126                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_1_b.npy"),
 127                   PadStrideInfo(1, 1, 1, 1))
 128               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 129               << ConvolutionLayer(
 130                   3U, 3U, 256U,
 131                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_2_w.npy"),
 132                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_2_b.npy"),
 133                   PadStrideInfo(1, 1, 1, 1))
 134               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 135               << ConvolutionLayer(
 136                   3U, 3U, 256U,
 137                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_3_w.npy"),
 138                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_3_b.npy"),
 139                   PadStrideInfo(1, 1, 1, 1))
 140               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 141               << ConvolutionLayer(
 142                   3U, 3U, 256U,
 143                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_4_w.npy"),
 144                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_4_b.npy"),
 145                   PadStrideInfo(1, 1, 1, 1))
 146               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 147               << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
 148               // Layer 4
 149               << ConvolutionLayer(
 150                   3U, 3U, 512U,
 151                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_1_w.npy"),
 152                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_1_b.npy"),
 153                   PadStrideInfo(1, 1, 1, 1))
 154               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 155               << ConvolutionLayer(
 156                   3U, 3U, 512U,
 157                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_2_w.npy"),
 158                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_2_b.npy"),
 159                   PadStrideInfo(1, 1, 1, 1))
 160               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 161               << ConvolutionLayer(
 162                   3U, 3U, 512U,
 163                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_3_w.npy"),
 164                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_3_b.npy"),
 165                   PadStrideInfo(1, 1, 1, 1))
 166               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 167               << ConvolutionLayer(
 168                   3U, 3U, 512U,
 169                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_4_w.npy"),
 170                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_4_b.npy"),
 171                   PadStrideInfo(1, 1, 1, 1))
 172               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 173               << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
 174               // Layer 5
 175               << ConvolutionLayer(
 176                   3U, 3U, 512U,
 177                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_1_w.npy"),
 178                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_1_b.npy"),
 179                   PadStrideInfo(1, 1, 1, 1))
 180               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 181               << ConvolutionLayer(
 182                   3U, 3U, 512U,
 183                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_2_w.npy"),
 184                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_2_b.npy"),
 185                   PadStrideInfo(1, 1, 1, 1))
 186               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 187               << ConvolutionLayer(
 188                   3U, 3U, 512U,
 189                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_3_w.npy"),
 190                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_3_b.npy"),
 191                   PadStrideInfo(1, 1, 1, 1))
 192               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 193               << ConvolutionLayer(
 194                   3U, 3U, 512U,
 195                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_4_w.npy"),
 196                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_4_b.npy"),
 197                   PadStrideInfo(1, 1, 1, 1))
 198               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 199               << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, PadStrideInfo(2, 2, 0, 0)))
 200               // Layer 6
 201               << FullyConnectedLayer(
 202                   4096U,
 203                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc6_w.npy"),
 204                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc6_b.npy"))
 205               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 206               // Layer 7
 207               << FullyConnectedLayer(
 208                   4096U,
 209                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc7_w.npy"),
 210                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc7_b.npy"))
 211               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
 212               // Layer 8
 213               << FullyConnectedLayer(
 214                   1000U,
 215                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc8_w.npy"),
 216                   get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc8_b.npy"))
 217               // Softmax
 218               << SoftmaxLayer()
 219               << Tensor(get_output_accessor(label, 5));
 220     }
 221     void do_run() override
 222     {
 223         // Run graph
 224         graph.run();
 225     }
 226
 227 private:
 228     Graph graph{};
 229 };
 230
 231 /** Main program for VGG19
 232  *
 233  * @param[in] argc Number of arguments
 234  * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL), [optional] Path to the weights folder, [optional] image, [optional] labels )
 235  */
 236 int main(int argc, char **argv)
 237 {
 238     return arm_compute::utils::run_example<GraphVGG19Example>(argc, argv);
 239 }