1 /*
2 * Copyright (c) 2017-2021 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/CommonGraphOptions.h"
27 #include "utils/GraphUtils.h"
28 #include "utils/Utils.h"
29
30 using namespace arm_compute::utils;
31 using namespace arm_compute::graph::frontend;
32 using namespace arm_compute::graph_utils;
33
34 /** Example demonstrating how to implement Googlenet's network using the Compute Library's graph API */
35 class GraphGooglenetExample : public Example
36 {
37 public:
GraphGooglenetExample()38 GraphGooglenetExample()
39 : cmd_parser(), common_opts(cmd_parser), common_params(), graph(0, "GoogleNet")
40 {
41 }
do_setup(int argc,char ** argv)42 bool do_setup(int argc, char **argv) override
43 {
44 // Parse arguments
45 cmd_parser.parse(argc, argv);
46 cmd_parser.validate();
47
48 // Consume common parameters
49 common_params = consume_common_graph_parameters(common_opts);
50
51 // Return when help menu is requested
52 if(common_params.help)
53 {
54 cmd_parser.print_help(argv[0]);
55 return false;
56 }
57
58 // Checks
59 ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "QASYMM8 not supported for this graph");
60
61 // Print parameter values
62 std::cout << common_params << std::endl;
63
64 // Get trainable parameters data path
65 std::string data_path = common_params.data_path;
66
67 // Create a preprocessor object
68 const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
69 std::unique_ptr<IPreprocessor> preprocessor = std::make_unique<CaffePreproccessor>(mean_rgb);
70
71 // Create input descriptor
72 const auto operation_layout = common_params.data_layout;
73 const TensorShape tensor_shape = permute_shape(TensorShape(224U, 224U, 3U, common_params.batches), DataLayout::NCHW, operation_layout);
74 TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(operation_layout);
75
76 // Set weights trained layout
77 const DataLayout weights_layout = DataLayout::NCHW;
78
79 graph << common_params.target
80 << common_params.fast_math_hint
81 << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor)))
82 << ConvolutionLayer(
83 7U, 7U, 64U,
84 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv1/conv1_7x7_s2_w.npy", weights_layout),
85 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv1/conv1_7x7_s2_b.npy"),
86 PadStrideInfo(2, 2, 3, 3))
87 .set_name("conv1/7x7_s2")
88 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv1/relu_7x7")
89 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool1/3x3_s2")
90 << NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f)).set_name("pool1/norm1")
91 << ConvolutionLayer(
92 1U, 1U, 64U,
93 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_reduce_w.npy", weights_layout),
94 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_reduce_b.npy"),
95 PadStrideInfo(1, 1, 0, 0))
96 .set_name("conv2/3x3_reduce")
97 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv2/relu_3x3_reduce")
98 << ConvolutionLayer(
99 3U, 3U, 192U,
100 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_w.npy", weights_layout),
101 get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_b.npy"),
102 PadStrideInfo(1, 1, 1, 1))
103 .set_name("conv2/3x3")
104 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv2/relu_3x3")
105 << NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f)).set_name("conv2/norm2")
106 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool2/3x3_s2");
107 graph << get_inception_node(data_path, "inception_3a", weights_layout, 64, std::make_tuple(96U, 128U), std::make_tuple(16U, 32U), 32U).set_name("inception_3a/concat");
108 graph << get_inception_node(data_path, "inception_3b", weights_layout, 128, std::make_tuple(128U, 192U), std::make_tuple(32U, 96U), 64U).set_name("inception_3b/concat");
109 graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool3/3x3_s2");
110 graph << get_inception_node(data_path, "inception_4a", weights_layout, 192, std::make_tuple(96U, 208U), std::make_tuple(16U, 48U), 64U).set_name("inception_4a/concat");
111 graph << get_inception_node(data_path, "inception_4b", weights_layout, 160, std::make_tuple(112U, 224U), std::make_tuple(24U, 64U), 64U).set_name("inception_4b/concat");
112 graph << get_inception_node(data_path, "inception_4c", weights_layout, 128, std::make_tuple(128U, 256U), std::make_tuple(24U, 64U), 64U).set_name("inception_4c/concat");
113 graph << get_inception_node(data_path, "inception_4d", weights_layout, 112, std::make_tuple(144U, 288U), std::make_tuple(32U, 64U), 64U).set_name("inception_4d/concat");
114 graph << get_inception_node(data_path, "inception_4e", weights_layout, 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U).set_name("inception_4e/concat");
115 graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool4/3x3_s2");
116 graph << get_inception_node(data_path, "inception_5a", weights_layout, 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U).set_name("inception_5a/concat");
117 graph << get_inception_node(data_path, "inception_5b", weights_layout, 384, std::make_tuple(192U, 384U), std::make_tuple(48U, 128U), 128U).set_name("inception_5b/concat");
118 graph << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 7, operation_layout, PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL))).set_name("pool5/7x7_s1")
119 << FullyConnectedLayer(
120 1000U,
121 get_weights_accessor(data_path, "/cnn_data/googlenet_model/loss3/loss3_classifier_w.npy", weights_layout),
122 get_weights_accessor(data_path, "/cnn_data/googlenet_model/loss3/loss3_classifier_b.npy"))
123 .set_name("loss3/classifier")
124 << SoftmaxLayer().set_name("prob")
125 << OutputLayer(get_output_accessor(common_params, 5));
126
127 // Finalize graph
128 GraphConfig config;
129 config.num_threads = common_params.threads;
130 config.use_tuner = common_params.enable_tuner;
131 config.tuner_mode = common_params.tuner_mode;
132 config.tuner_file = common_params.tuner_file;
133 config.mlgo_file = common_params.mlgo_file;
134
135 graph.finalize(common_params.target, config);
136
137 return true;
138 }
do_run()139 void do_run() override
140 {
141 // Run graph
142 graph.run();
143 }
144
145 private:
146 CommandLineParser cmd_parser;
147 CommonGraphOptions common_opts;
148 CommonGraphParams common_params;
149 Stream graph;
150
get_inception_node(const std::string & data_path,std::string && param_path,DataLayout weights_layout,unsigned int a_filt,std::tuple<unsigned int,unsigned int> b_filters,std::tuple<unsigned int,unsigned int> c_filters,unsigned int d_filt)151 ConcatLayer get_inception_node(const std::string &data_path, std::string &¶m_path, DataLayout weights_layout,
152 unsigned int a_filt,
153 std::tuple<unsigned int, unsigned int> b_filters,
154 std::tuple<unsigned int, unsigned int> c_filters,
155 unsigned int d_filt)
156 {
157 std::string total_path = "/cnn_data/googlenet_model/" + param_path + "/" + param_path + "_";
158 SubStream i_a(graph);
159 i_a << ConvolutionLayer(
160 1U, 1U, a_filt,
161 get_weights_accessor(data_path, total_path + "1x1_w.npy", weights_layout),
162 get_weights_accessor(data_path, total_path + "1x1_b.npy"),
163 PadStrideInfo(1, 1, 0, 0))
164 .set_name(param_path + "/1x1")
165 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_1x1");
166
167 SubStream i_b(graph);
168 i_b << ConvolutionLayer(
169 1U, 1U, std::get<0>(b_filters),
170 get_weights_accessor(data_path, total_path + "3x3_reduce_w.npy", weights_layout),
171 get_weights_accessor(data_path, total_path + "3x3_reduce_b.npy"),
172 PadStrideInfo(1, 1, 0, 0))
173 .set_name(param_path + "/3x3_reduce")
174 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_3x3_reduce")
175 << ConvolutionLayer(
176 3U, 3U, std::get<1>(b_filters),
177 get_weights_accessor(data_path, total_path + "3x3_w.npy", weights_layout),
178 get_weights_accessor(data_path, total_path + "3x3_b.npy"),
179 PadStrideInfo(1, 1, 1, 1))
180 .set_name(param_path + "/3x3")
181 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_3x3");
182
183 SubStream i_c(graph);
184 i_c << ConvolutionLayer(
185 1U, 1U, std::get<0>(c_filters),
186 get_weights_accessor(data_path, total_path + "5x5_reduce_w.npy", weights_layout),
187 get_weights_accessor(data_path, total_path + "5x5_reduce_b.npy"),
188 PadStrideInfo(1, 1, 0, 0))
189 .set_name(param_path + "/5x5_reduce")
190 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_5x5_reduce")
191 << ConvolutionLayer(
192 5U, 5U, std::get<1>(c_filters),
193 get_weights_accessor(data_path, total_path + "5x5_w.npy", weights_layout),
194 get_weights_accessor(data_path, total_path + "5x5_b.npy"),
195 PadStrideInfo(1, 1, 2, 2))
196 .set_name(param_path + "/5x5")
197 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_5x5");
198
199 SubStream i_d(graph);
200 i_d << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, common_params.data_layout, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL))).set_name(param_path + "/pool")
201 << ConvolutionLayer(
202 1U, 1U, d_filt,
203 get_weights_accessor(data_path, total_path + "pool_proj_w.npy", weights_layout),
204 get_weights_accessor(data_path, total_path + "pool_proj_b.npy"),
205 PadStrideInfo(1, 1, 0, 0))
206 .set_name(param_path + "/pool_proj")
207 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_pool_proj");
208
209 return ConcatLayer(std::move(i_a), std::move(i_b), std::move(i_c), std::move(i_d));
210 }
211 };
212
213 /** Main program for Googlenet
214 *
215 * Model is based on:
216 * https://arxiv.org/abs/1409.4842
217 * "Going deeper with convolutions"
218 * Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich
219 *
220 * Provenance: https://github.com/BVLC/caffe/tree/master/models/bvlc_googlenet
221 *
222 * @note To list all the possible arguments execute the binary appended with the --help option
223 *
224 * @param[in] argc Number of arguments
225 * @param[in] argv Arguments
226 */
main(int argc,char ** argv)227 int main(int argc, char **argv)
228 {
229 return arm_compute::utils::run_example<GraphGooglenetExample>(argc, argv);
230 }
231