2 * Copyright (c) 2017-2018 ARM Limited.
4 * SPDX-License-Identifier: MIT
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:
13 * The above copyright notice and this permission notice shall be included in all
14 * copies or substantial portions of the Software.
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
24 #include "arm_compute/graph.h"
25 #include "support/ToolchainSupport.h"
26 #include "utils/GraphUtils.h"
27 #include "utils/Utils.h"
32 using namespace arm_compute::utils;
33 using namespace arm_compute::graph::frontend;
34 using namespace arm_compute::graph_utils;
36 /** Example demonstrating how to implement Googlenet's network using the Compute Library's graph API
38 * @param[in] argc Number of arguments
39 * @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 class GraphGooglenetExample : public Example
44 void do_setup(int argc, char **argv) override
46 std::string data_path; /* Path to the trainable data */
47 std::string image; /* Image data */
48 std::string label; /* Label data */
50 // Create a preprocessor object
51 const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
52 std::unique_ptr<IPreprocessor> preprocessor = arm_compute::support::cpp14::make_unique<CaffePreproccessor>(mean_rgb);
54 // Set target. 0 (NEON), 1 (OpenCL), 2 (OpenCL with Tuner). By default it is NEON
55 const int target = argc > 1 ? std::strtol(argv[1], nullptr, 10) : 0;
56 Target target_hint = set_target_hint(target);
57 FastMathHint fast_math_hint = FastMathHint::DISABLED;
63 std::cout << "Usage: " << argv[0] << " [target] [path_to_data] [image] [labels] [fast_math_hint]\n\n";
64 std::cout << "No data folder provided: using random values\n\n";
68 std::cout << "Usage: " << argv[0] << " " << argv[1] << " [path_to_data] [image] [labels] [fast_math_hint]\n\n";
69 std::cout << "No data folder provided: using random values\n\n";
74 std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " [image] [labels] [fast_math_hint]\n\n";
75 std::cout << "No image provided: using random values\n\n";
81 std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " [labels] [fast_math_hint]\n\n";
82 std::cout << "No text file with labels provided: skipping output accessor\n\n";
89 std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " " << argv[4] << " [fast_math_hint]\n\n";
90 std::cout << "No fast math info provided: disabling fast math\n\n";
97 fast_math_hint = (std::strtol(argv[5], nullptr, 1) == 0) ? FastMathHint::DISABLED : FastMathHint::ENABLED;
102 << InputLayer(TensorDescriptor(TensorShape(224U, 224U, 3U, 1U), DataType::F32),
103 get_input_accessor(image, std::move(preprocessor)))
106 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv1/conv1_7x7_s2_w.npy"),
107 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv1/conv1_7x7_s2_b.npy"),
108 PadStrideInfo(2, 2, 3, 3))
109 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
110 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
111 << NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f))
114 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_reduce_w.npy"),
115 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_reduce_b.npy"),
116 PadStrideInfo(1, 1, 0, 0))
117 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
120 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_w.npy"),
121 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_b.npy"),
122 PadStrideInfo(1, 1, 1, 1))
123 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
124 << NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f))
125 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)));
126 graph << get_inception_node(data_path, "inception_3a", 64, std::make_tuple(96U, 128U), std::make_tuple(16U, 32U), 32U);
127 graph << get_inception_node(data_path, "inception_3b", 128, std::make_tuple(128U, 192U), std::make_tuple(32U, 96U), 64U);
128 graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)));
129 graph << get_inception_node(data_path, "inception_4a", 192, std::make_tuple(96U, 208U), std::make_tuple(16U, 48U), 64U);
130 graph << get_inception_node(data_path, "inception_4b", 160, std::make_tuple(112U, 224U), std::make_tuple(24U, 64U), 64U);
131 graph << get_inception_node(data_path, "inception_4c", 128, std::make_tuple(128U, 256U), std::make_tuple(24U, 64U), 64U);
132 graph << get_inception_node(data_path, "inception_4d", 112, std::make_tuple(144U, 288U), std::make_tuple(32U, 64U), 64U);
133 graph << get_inception_node(data_path, "inception_4e", 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U);
134 graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)));
135 graph << get_inception_node(data_path, "inception_5a", 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U);
136 graph << get_inception_node(data_path, "inception_5b", 384, std::make_tuple(192U, 384U), std::make_tuple(48U, 128U), 128U);
137 graph << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 7, PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL)))
138 << FullyConnectedLayer(
140 get_weights_accessor(data_path, "/cnn_data/googlenet_model/loss3/loss3_classifier_w.npy"),
141 get_weights_accessor(data_path, "/cnn_data/googlenet_model/loss3/loss3_classifier_b.npy"))
143 << OutputLayer(get_output_accessor(label, 5));
147 config.use_tuner = (target == 2);
148 graph.finalize(target_hint, config);
150 void do_run() override
157 Stream graph{ 0, "GoogleNet" };
159 BranchLayer get_inception_node(const std::string &data_path, std::string &¶m_path,
161 std::tuple<unsigned int, unsigned int> b_filters,
162 std::tuple<unsigned int, unsigned int> c_filters,
165 std::string total_path = "/cnn_data/googlenet_model/" + param_path + "/" + param_path + "_";
166 SubStream i_a(graph);
167 i_a << ConvolutionLayer(
169 get_weights_accessor(data_path, total_path + "1x1_w.npy"),
170 get_weights_accessor(data_path, total_path + "1x1_b.npy"),
171 PadStrideInfo(1, 1, 0, 0))
172 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
174 SubStream i_b(graph);
175 i_b << ConvolutionLayer(
176 1U, 1U, std::get<0>(b_filters),
177 get_weights_accessor(data_path, total_path + "3x3_reduce_w.npy"),
178 get_weights_accessor(data_path, total_path + "3x3_reduce_b.npy"),
179 PadStrideInfo(1, 1, 0, 0))
180 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
182 3U, 3U, std::get<1>(b_filters),
183 get_weights_accessor(data_path, total_path + "3x3_w.npy"),
184 get_weights_accessor(data_path, total_path + "3x3_b.npy"),
185 PadStrideInfo(1, 1, 1, 1))
186 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
188 SubStream i_c(graph);
189 i_c << ConvolutionLayer(
190 1U, 1U, std::get<0>(c_filters),
191 get_weights_accessor(data_path, total_path + "5x5_reduce_w.npy"),
192 get_weights_accessor(data_path, total_path + "5x5_reduce_b.npy"),
193 PadStrideInfo(1, 1, 0, 0))
194 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
196 5U, 5U, std::get<1>(c_filters),
197 get_weights_accessor(data_path, total_path + "5x5_w.npy"),
198 get_weights_accessor(data_path, total_path + "5x5_b.npy"),
199 PadStrideInfo(1, 1, 2, 2))
200 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
202 SubStream i_d(graph);
203 i_d << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL)))
206 get_weights_accessor(data_path, total_path + "pool_proj_w.npy"),
207 get_weights_accessor(data_path, total_path + "pool_proj_b.npy"),
208 PadStrideInfo(1, 1, 0, 0))
209 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
211 return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b), std::move(i_c), std::move(i_d));
215 /** Main program for Googlenet
217 * @param[in] argc Number of arguments
218 * @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) )
220 int main(int argc, char **argv)
222 return arm_compute::utils::run_example<GraphGooglenetExample>(argc, argv);