Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / ComputeLibrary / 8140e1e155d3430992fa46e04ef8938ff09ffd2d / . / examples / graph_squeezenet.cpp
blob: c8c411aa8bff500389705dcdf9644fcd2b99b3d8 [file] [log] [blame]
Anthony Barbier8140e1e2017-12-14 23:48:46 +0000[diff] [blame^]1/*
2 * Copyright (c) 2017 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
34using namespace arm_compute::graph;
35using namespace arm_compute::graph_utils;
36using namespace arm_compute::logging;
37
38namespace
39{
40BranchLayer get_expand_fire_node(const std::string &data_path, std::string &&param_path, unsigned int expand1_filt, unsigned int expand3_filt)
41{
42 std::string total_path = "/cnn_data/squeezenet_v1.0_model/" + param_path + "_";
43 SubGraph i_a;
44 i_a << ConvolutionLayer(
45 1U, 1U, expand1_filt,
46 get_weights_accessor(data_path, total_path + "expand1x1_w.npy"),
47 get_weights_accessor(data_path, total_path + "expand1x1_b.npy"),
48 PadStrideInfo(1, 1, 0, 0))
49 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
50
51 SubGraph i_b;
52 i_b << ConvolutionLayer(
53 3U, 3U, expand3_filt,
54 get_weights_accessor(data_path, total_path + "expand3x3_w.npy"),
55 get_weights_accessor(data_path, total_path + "expand3x3_b.npy"),
56 PadStrideInfo(1, 1, 1, 1))
57 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
58
59 return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b));
60}
61} // namespace
62
63/** Example demonstrating how to implement Squeezenet's network using the Compute Library's graph API
64 *
65 * @param[in] argc Number of arguments
66 * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL), [optional] Path to the weights folder, [optional] image, [optional] labels )
67 */
68void main_graph_squeezenet(int argc, const char **argv)
69{
70 std::string data_path; /* Path to the trainable data */
71 std::string image; /* Image data */
72 std::string label; /* Label data */
73
74 constexpr float mean_r = 122.68f; /* Mean value to subtract from red channel */
75 constexpr float mean_g = 116.67f; /* Mean value to subtract from green channel */
76 constexpr float mean_b = 104.01f; /* Mean value to subtract from blue channel */
77
78 // Set target. 0 (NEON), 1 (OpenCL). By default it is NEON
79 TargetHint target_hint = set_target_hint(argc > 1 ? std::strtol(argv[1], nullptr, 10) : 0);
80 ConvolutionMethodHint convolution_hint = target_hint == TargetHint::NEON ? ConvolutionMethodHint::GEMM : ConvolutionMethodHint::DIRECT;
81
82 // Parse arguments
83 if(argc < 2)
84 {
85 // Print help
86 std::cout << "Usage: " << argv[0] << " [target] [path_to_data] [image] [labels]\n\n";
87 std::cout << "No data folder provided: using random values\n\n";
88 }
89 else if(argc == 2)
90 {
91 std::cout << "Usage: " << argv[0] << " " << argv[1] << " [path_to_data] [image] [labels]\n\n";
92 std::cout << "No data folder provided: using random values\n\n";
93 }
94 else if(argc == 3)
95 {
96 data_path = argv[2];
97 std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " [image] [labels]\n\n";
98 std::cout << "No image provided: using random values\n\n";
99 }
100 else if(argc == 4)
101 {
102 data_path = argv[2];
103 image = argv[3];
104 std::cout << "Usage: " << argv[0] << " " << argv[1] << " " << argv[2] << " " << argv[3] << " [labels]\n\n";
105 std::cout << "No text file with labels provided: skipping output accessor\n\n";
106 }
107 else
108 {
109 data_path = argv[2];
110 image = argv[3];
111 label = argv[4];
112 }
113
114 Graph graph;
115
116 graph << target_hint
117 << Tensor(TensorInfo(TensorShape(224U, 224U, 3U, 1U), 1, DataType::F32),
118 get_input_accessor(image, mean_r, mean_g, mean_b))
119 << ConvolutionLayer(
120 7U, 7U, 96U,
121 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv1_w.npy"),
122 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv1_b.npy"),
123 PadStrideInfo(2, 2, 0, 0))
124 << convolution_hint
125 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
126 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
127 << ConvolutionLayer(
128 1U, 1U, 16U,
129 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire2_squeeze1x1_w.npy"),
130 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire2_squeeze1x1_b.npy"),
131 PadStrideInfo(1, 1, 0, 0))
132 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
133 << get_expand_fire_node(data_path, "fire2", 64U, 64U)
134 << ConvolutionLayer(
135 1U, 1U, 16U,
136 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire3_squeeze1x1_w.npy"),
137 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire3_squeeze1x1_b.npy"),
138 PadStrideInfo(1, 1, 0, 0))
139 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
140 << get_expand_fire_node(data_path, "fire3", 64U, 64U)
141 << ConvolutionLayer(
142 1U, 1U, 32U,
143 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire4_squeeze1x1_w.npy"),
144 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire4_squeeze1x1_b.npy"),
145 PadStrideInfo(1, 1, 0, 0))
146 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
147 << get_expand_fire_node(data_path, "fire4", 128U, 128U)
148 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
149 << ConvolutionLayer(
150 1U, 1U, 32U,
151 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire5_squeeze1x1_w.npy"),
152 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire5_squeeze1x1_b.npy"),
153 PadStrideInfo(1, 1, 0, 0))
154 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
155 << get_expand_fire_node(data_path, "fire5", 128U, 128U)
156 << ConvolutionLayer(
157 1U, 1U, 48U,
158 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire6_squeeze1x1_w.npy"),
159 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire6_squeeze1x1_b.npy"),
160 PadStrideInfo(1, 1, 0, 0))
161 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
162 << get_expand_fire_node(data_path, "fire6", 192U, 192U)
163 << ConvolutionLayer(
164 1U, 1U, 48U,
165 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire7_squeeze1x1_w.npy"),
166 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire7_squeeze1x1_b.npy"),
167 PadStrideInfo(1, 1, 0, 0))
168 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
169 << get_expand_fire_node(data_path, "fire7", 192U, 192U)
170 << ConvolutionLayer(
171 1U, 1U, 64U,
172 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire8_squeeze1x1_w.npy"),
173 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire8_squeeze1x1_b.npy"),
174 PadStrideInfo(1, 1, 0, 0))
175 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
176 << get_expand_fire_node(data_path, "fire8", 256U, 256U)
177 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
178 << ConvolutionLayer(
179 1U, 1U, 64U,
180 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire9_squeeze1x1_w.npy"),
181 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire9_squeeze1x1_b.npy"),
182 PadStrideInfo(1, 1, 0, 0))
183 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
184 << get_expand_fire_node(data_path, "fire9", 256U, 256U)
185 << ConvolutionLayer(
186 1U, 1U, 1000U,
187 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv10_w.npy"),
188 get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv10_b.npy"),
189 PadStrideInfo(1, 1, 0, 0))
190 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
191 << PoolingLayer(PoolingLayerInfo(PoolingType::AVG))
192 << FlattenLayer()
193 << SoftmaxLayer()
194 << Tensor(get_output_accessor(label, 5));
195
196 graph.run();
197}
198
199/** Main program for Squeezenet v1.0
200 *
201 * @param[in] argc Number of arguments
202 * @param[in] argv Arguments ( [optional] Target (0 = NEON, 1 = OpenCL), [optional] Path to the weights folder, [optional] image, [optional] labels )
203 */
204int main(int argc, const char **argv)
205{
206 return arm_compute::utils::run_example(argc, argv, main_graph_squeezenet);
207}