Gitiles
Code Review Sign In
review.mlplatform.org / ml / ComputeLibrary / 4398becc0ae11f346e1669a9b5210ef389bd8012 / . / examples / graph_squeezenet.cpp
blob: 888dd5e5f1da4a26aa4b61da611e280f3b65a72c [file] [log] [blame]
Georgios Pinitas37561862017-10-19 10:51:03 +0100[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#ifndef ARM_COMPUTE_CL /* Needed by Utils.cpp to handle OpenCL exceptions properly */
25#error "This example needs to be built with -DARM_COMPUTE_CL"
26#endif /* ARM_COMPUTE_CL */
27
Isabella Gottardi4398bec2017-10-19 16:10:59 +0100[diff] [blame^]28#include "arm_compute/core/utils/logging/LoggerRegistry.h"
Georgios Pinitas37561862017-10-19 10:51:03 +0100[diff] [blame]29#include "arm_compute/graph/Graph.h"
30#include "arm_compute/graph/Nodes.h"
31#include "arm_compute/graph/SubGraph.h"
32#include "arm_compute/runtime/CL/CLScheduler.h"
33#include "arm_compute/runtime/Scheduler.h"
34#include "support/ToolchainSupport.h"
35#include "utils/GraphUtils.h"
36#include "utils/Utils.h"
37
38#include <cstdlib>
39#include <iostream>
40#include <memory>
41#include <tuple>
42
43using namespace arm_compute::graph;
44using namespace arm_compute::graph_utils;
Isabella Gottardi4398bec2017-10-19 16:10:59 +0100[diff] [blame^]45using namespace arm_compute::logging;
Georgios Pinitas37561862017-10-19 10:51:03 +0100[diff] [blame]46
47/** Generates appropriate accessor according to the specified path
48 *
49 * @note If path is empty will generate a DummyAccessor else will generate a NumPyBinLoader
50 *
51 * @param path Path to the data files
52 * @param data_file Relative path to the data files from path
53 *
54 * @return An appropriate tensor accessor
55 */
56std::unique_ptr<ITensorAccessor> get_accessor(const std::string &path, const std::string &data_file)
57{
58 if(path.empty())
59 {
60 return arm_compute::support::cpp14::make_unique<DummyAccessor>();
61 }
62 else
63 {
64 return arm_compute::support::cpp14::make_unique<NumPyBinLoader>(path + data_file);
65 }
66}
67
68BranchLayer get_expand_fire_node(const std::string &data_path, std::string &&param_path, unsigned int expand1_filt, unsigned int expand3_filt)
69{
70 std::string total_path = "/cnn_data/squeezenet_v1.0_model/" + param_path + "_";
71 SubGraph i_a;
72 i_a << ConvolutionLayer(
73 1U, 1U, expand1_filt,
74 get_accessor(data_path, total_path + "expand1x1_w.npy"),
75 get_accessor(data_path, total_path + "expand1x1_b.npy"),
76 PadStrideInfo(1, 1, 0, 0))
77 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
78
79 SubGraph i_b;
80 i_b << ConvolutionLayer(
81 3U, 3U, expand3_filt,
82 get_accessor(data_path, total_path + "expand3x3_w.npy"),
83 get_accessor(data_path, total_path + "expand3x3_b.npy"),
84 PadStrideInfo(1, 1, 1, 1))
85 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU));
86
87 return BranchLayer(BranchMergeMethod::DEPTH_CONCATENATE, std::move(i_a), std::move(i_b));
88}
89
90/** Example demonstrating how to implement Squeezenet's network using the Compute Library's graph API
91 *
92 * @param[in] argc Number of arguments
93 * @param[in] argv Arguments ( [optional] Path to the weights folder, [optional] batches )
94 */
95void main_graph_squeezenet(int argc, const char **argv)
96{
97 std::string data_path; /** Path to the trainable data */
98 unsigned int batches = 4; /** Number of batches */
99
100 // Parse arguments
101 if(argc < 2)
102 {
103 // Print help
104 std::cout << "Usage: " << argv[0] << " [path_to_data] [batches]\n\n";
105 std::cout << "No data folder provided: using random values\n\n";
106 }
107 else if(argc == 2)
108 {
109 //Do something with argv[1]
110 data_path = argv[1];
111 std::cout << "Usage: " << argv[0] << " [path_to_data] [batches]\n\n";
112 std::cout << "No number of batches where specified, thus will use the default : " << batches << "\n\n";
113 }
114 else
115 {
116 //Do something with argv[1] and argv[2]
117 data_path = argv[1];
118 batches = std::strtol(argv[2], nullptr, 0);
119 }
120
121 // Check if OpenCL is available and initialize the scheduler
122 if(arm_compute::opencl_is_available())
123 {
124 arm_compute::CLScheduler::get().default_init();
125 }
126
127 Graph graph;
Isabella Gottardi4398bec2017-10-19 16:10:59 +0100[diff] [blame^]128 LoggerRegistry::get().create_reserved_loggers(LogLevel::INFO, { std::make_shared<StdPrinter>() });
Georgios Pinitas37561862017-10-19 10:51:03 +0100[diff] [blame]129
130 graph << TargetHint::OPENCL
131 << Tensor(TensorInfo(TensorShape(224U, 224U, 3U, batches), 1, DataType::F32), DummyAccessor())
132 << ConvolutionLayer(
133 7U, 7U, 96U,
134 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv1_w.npy"),
135 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv1_b.npy"),
136 PadStrideInfo(2, 2, 0, 0))
137 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
138 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
139 << ConvolutionLayer(
140 1U, 1U, 16U,
141 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire2_squeeze1x1_w.npy"),
142 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire2_squeeze1x1_b.npy"),
143 PadStrideInfo(1, 1, 0, 0))
144 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
145 << get_expand_fire_node(data_path, "fire2", 64U, 64U)
146 << ConvolutionLayer(
147 1U, 1U, 16U,
148 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire3_squeeze1x1_w.npy"),
149 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire3_squeeze1x1_b.npy"),
150 PadStrideInfo(1, 1, 0, 0))
151 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
152 << get_expand_fire_node(data_path, "fire3", 64U, 64U)
153 << ConvolutionLayer(
154 1U, 1U, 32U,
155 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire4_squeeze1x1_w.npy"),
156 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire4_squeeze1x1_b.npy"),
157 PadStrideInfo(1, 1, 0, 0))
158 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
159 << get_expand_fire_node(data_path, "fire4", 128U, 128U)
160 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
161 << ConvolutionLayer(
162 1U, 1U, 32U,
163 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire5_squeeze1x1_w.npy"),
164 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire5_squeeze1x1_b.npy"),
165 PadStrideInfo(1, 1, 0, 0))
166 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
167 << get_expand_fire_node(data_path, "fire5", 128U, 128U)
168 << ConvolutionLayer(
169 1U, 1U, 48U,
170 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire6_squeeze1x1_w.npy"),
171 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire6_squeeze1x1_b.npy"),
172 PadStrideInfo(1, 1, 0, 0))
173 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
174 << get_expand_fire_node(data_path, "fire6", 192U, 192U)
175 << ConvolutionLayer(
176 1U, 1U, 48U,
177 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire7_squeeze1x1_w.npy"),
178 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire7_squeeze1x1_b.npy"),
179 PadStrideInfo(1, 1, 0, 0))
180 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
181 << get_expand_fire_node(data_path, "fire7", 192U, 192U)
182 << ConvolutionLayer(
183 1U, 1U, 64U,
184 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire8_squeeze1x1_w.npy"),
185 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire8_squeeze1x1_b.npy"),
186 PadStrideInfo(1, 1, 0, 0))
187 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
188 << get_expand_fire_node(data_path, "fire8", 256U, 256U)
189 << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL)))
190 << ConvolutionLayer(
191 1U, 1U, 64U,
192 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire9_squeeze1x1_w.npy"),
193 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire9_squeeze1x1_b.npy"),
194 PadStrideInfo(1, 1, 0, 0))
195 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
196 << get_expand_fire_node(data_path, "fire9", 256U, 256U)
197 << ConvolutionLayer(
198 1U, 1U, 1000U,
199 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv10_w.npy"),
200 get_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv10_b.npy"),
201 PadStrideInfo(1, 1, 0, 0))
202 << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU))
203 << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 13, PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL)))
204 << SoftmaxLayer()
205 << Tensor(DummyAccessor());
206
207 graph.run();
208}
209
210/** Main program for Squeezenet v1.0
211 *
212 * @param[in] argc Number of arguments
213 * @param[in] argv Arguments ( [optional] Path to the weights folder, [optional] batches )
214 */
215int main(int argc, const char **argv)
216{
217 return arm_compute::utils::run_example(argc, argv, main_graph_squeezenet);
218}