examples/graph_alexnet.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.h"
  25 #include "support/ToolchainSupport.h"
  26 #include "utils/GraphUtils.h"
  27 #include "utils/Utils.h"
  28
  29 #include <cstdlib>
  30 #include <iostream>
  31 #include <memory>
  32
  33 using namespace arm_compute::utils;
  34 using namespace arm_compute::graph::frontend;
  35 using namespace arm_compute::graph_utils;
  36
  37 /** Example demonstrating how to implement AlexNet's network using the Compute Library's graph API
  38  *
  39  * @param[in] argc Number of arguments
  40  * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL, 2 = OpenCL with Tuner), [optional] Path to the weights folder, [optional] image, [optional] labels, [optional] Fast math for convolution layer (0 = DISABLED, 1 = ENABLED) )
  41  */
  42 class GraphAlexnetExample : public Example
  43 {
  44 public:
  45     void do_setup(int argc, char **argv) override
  46     {
  47         std::string data_path; /* Path to the trainable data */
  48         std::string image;     /* Image data */
  49         std::string label;     /* Label data */
  50
  51         // Create a preprocessor object
  52         const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
  53         std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb);
  54
  55         // Set target. 0 (NEON), 1 (OpenCL), 2 (OpenCL with Tuner). By default it is NEON
  56         const int target      = argc > 1 ? std::strtol(argv[1], nullptr, 10) : 0;
  57         Target    target_hint = set_target_hint(target);
  58
  59         const bool        is_neon              = (target_hint == Target::NEON);
  60         ConvolutionMethod convolution_5x5_hint = is_neon ? ConvolutionMethod::GEMM : ConvolutionMethod::DIRECT;
  61         ConvolutionMethod convolution_3x3_hint = ConvolutionMethod::DEFAULT;
  62         FastMathHint      fast_math_hint       = FastMathHint::DISABLED;
  63
  64         // Parse arguments
  65         if(argc < 2)
  66         {
  67             // Print help
  68             std::cout << "Usage: " << argv[0] << " [target] [path_to_data] [image] [labels] [fast_math_hint]\n\n";
  69             std::cout << "No data folder provided: using random values\n\n";
  70         }
  71         else if(argc == 2)
  72         {
  73             std::cout << "Usage: " << argv[0] << " " << argv[1] << " [path_to_data] [image] [labels] [fast_math_hint]\n\n";
  74             std::cout << "No data folder provided: using random values\n\n";
  75         }
  76         else if(argc == 3)
  77         {
  78             data_path = argv[2];
  79             std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " [image] [labels] [fast_math_hint]\n\n";
  80             std::cout << "No image provided: using random values\n\n";
  81         }
  82         else if(argc == 4)
  83         {
  84             data_path = argv[2];
  85             image     = argv[3];
  86             std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " [labels] [fast_math_hint]\n\n";
  87             std::cout << "No text file with labels provided: skipping output accessor\n\n";
  88         }
  89         else if(argc == 5)
  90         {
  91             data_path = argv[2];
  92             image     = argv[3];
  93             label     = argv[4];
  94             std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " " << argv[4] << " [fast_math_hint]\n\n";
  95             std::cout << "No fast math info provided: disabling fast math\n\n";
  96         }
  97         else
  98         {
  99             data_path      = argv[2];
 100             image          = argv[3];
 101             label          = argv[4];
 102             fast_math_hint = (std::strtol(argv[5], nullptr, 1) == 0) ? FastMathHint::DISABLED : FastMathHint::ENABLED;
 103         }
 104
 105         graph << target_hint
 106               << fast_math_hint
 107               << InputLayer(TensorDescriptor(TensorShape(227U, 227U, 3U, 1U), DataType::F32),
 108                             get_input_accessor(image, std::move(preprocessor)))
 109               // Layer 1
 110               << ConvolutionLayer(
 111                   11U, 11U, 96U,
 112                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv1_w.npy"),
 113                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv1_b.npy"),
 114                   PadStrideInfo(4, 4, 0, 0))
 115               .set_name("conv1")
 116               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu1")
 117               << NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f)).set_name("norm1")
 118               << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0))).set_name("pool1")
 119               // Layer 2
 120               << convolution_5x5_hint
 121               << ConvolutionLayer(
 122                   5U, 5U, 256U,
 123                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv2_w.npy"),
 124                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv2_b.npy"),
 125                   PadStrideInfo(1, 1, 2, 2), 2)
 126               .set_name("conv2")
 127               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu2")
 128               << NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f)).set_name("norm2")
 129               << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0))).set_name("pool2")
 130               << convolution_3x3_hint
 131               // Layer 3
 132               << ConvolutionLayer(
 133                   3U, 3U, 384U,
 134                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv3_w.npy"),
 135                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv3_b.npy"),
 136                   PadStrideInfo(1, 1, 1, 1))
 137               .set_name("conv3")
 138               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu3")
 139               // Layer 4
 140               << ConvolutionLayer(
 141                   3U, 3U, 384U,
 142                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv4_w.npy"),
 143                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv4_b.npy"),
 144                   PadStrideInfo(1, 1, 1, 1), 2)
 145               .set_name("conv4")
 146               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu4")
 147               // Layer 5
 148               << ConvolutionLayer(
 149                   3U, 3U, 256U,
 150                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv5_w.npy"),
 151                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/conv5_b.npy"),
 152                   PadStrideInfo(1, 1, 1, 1), 2)
 153               .set_name("conv5")
 154               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu5")
 155               << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0))).set_name("pool5")
 156               // Layer 6
 157               << FullyConnectedLayer(
 158                   4096U,
 159                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc6_w.npy"),
 160                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc6_b.npy"))
 161               .set_name("fc6")
 162               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu6")
 163               // Layer 7
 164               << FullyConnectedLayer(
 165                   4096U,
 166                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc7_w.npy"),
 167                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc7_b.npy"))
 168               .set_name("fc7")
 169               << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu7")
 170               // Layer 8
 171               << FullyConnectedLayer(
 172                   1000U,
 173                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc8_w.npy"),
 174                   get_weights_accessor(data_path, "/cnn_data/alexnet_model/fc8_b.npy"))
 175               .set_name("fc8")
 176               // Softmax
 177               << SoftmaxLayer().set_name("prob")
 178               << OutputLayer(get_output_accessor(label, 5));
 179
 180         // Finalize graph
 181         GraphConfig config;
 182         config.use_tuner = (target == 2);
 183         graph.finalize(target_hint, config);
 184     }
 185     void do_run() override
 186     {
 187         // Run graph
 188         graph.run();
 189     }
 190
 191 private:
 192     Stream graph{ 0, "AlexNet" };
 193 };
 194
 195 /** Main program for AlexNet
 196  *
 197  * @param[in] argc Number of arguments
 198  * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL, 2 = OpenCL with Tuner), [optional] Path to the weights folder, [optional] image, [optional] labels, [optional] Fast math for convolution layer (0 = DISABLED, 1 = ENABLED) )
 199  */
 200 int main(int argc, char **argv)
 201 {
 202     return arm_compute::utils::run_example<GraphAlexnetExample>(argc, argv);
 203 }