Compute Library
 21.02
graph_squeezenet.cpp
Go to the documentation of this file.
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"
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 Squeezenet's network using the Compute Library's graph API */
35 class GraphSqueezenetExample : public Example
36 {
37 public:
38  GraphSqueezenetExample()
39  : cmd_parser(), common_opts(cmd_parser), common_params(), graph(0, "SqueezeNetV1")
40  {
41  }
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  // Print parameter values
59  std::cout << common_params << std::endl;
60 
61  // Get trainable parameters data path
62  std::string data_path = common_params.data_path;
63 
64  // Create a preprocessor object
65  const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
66  std::unique_ptr<IPreprocessor> preprocessor = std::make_unique<CaffePreproccessor>(mean_rgb);
67 
68  // Create input descriptor
69  const auto operation_layout = common_params.data_layout;
70  const TensorShape tensor_shape = permute_shape(TensorShape(224U, 224U, 3U, 1U), DataLayout::NCHW, operation_layout);
71  TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(operation_layout);
72 
73  // Set weights trained layout
74  const DataLayout weights_layout = DataLayout::NCHW;
75 
76  graph << common_params.target
77  << common_params.fast_math_hint
78  << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor)))
80  7U, 7U, 96U,
81  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv1_w.npy", weights_layout),
82  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv1_b.npy"),
83  PadStrideInfo(2, 2, 0, 0))
84  .set_name("conv1")
85  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu_conv1")
86  << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool1")
88  1U, 1U, 16U,
89  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire2_squeeze1x1_w.npy", weights_layout),
90  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire2_squeeze1x1_b.npy"),
91  PadStrideInfo(1, 1, 0, 0))
92  .set_name("fire2/squeeze1x1")
93  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("fire2/relu_squeeze1x1");
94  graph << get_expand_fire_node(data_path, "fire2", weights_layout, 64U, 64U).set_name("fire2/concat");
95  graph << ConvolutionLayer(
96  1U, 1U, 16U,
97  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire3_squeeze1x1_w.npy", weights_layout),
98  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire3_squeeze1x1_b.npy"),
99  PadStrideInfo(1, 1, 0, 0))
100  .set_name("fire3/squeeze1x1")
101  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("fire3/relu_squeeze1x1");
102  graph << get_expand_fire_node(data_path, "fire3", weights_layout, 64U, 64U).set_name("fire3/concat");
103  graph << ConvolutionLayer(
104  1U, 1U, 32U,
105  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire4_squeeze1x1_w.npy", weights_layout),
106  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire4_squeeze1x1_b.npy"),
107  PadStrideInfo(1, 1, 0, 0))
108  .set_name("fire4/squeeze1x1")
109  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("fire4/relu_squeeze1x1");
110  graph << get_expand_fire_node(data_path, "fire4", weights_layout, 128U, 128U).set_name("fire4/concat");
111  graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool4")
112  << ConvolutionLayer(
113  1U, 1U, 32U,
114  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire5_squeeze1x1_w.npy", weights_layout),
115  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire5_squeeze1x1_b.npy"),
116  PadStrideInfo(1, 1, 0, 0))
117  .set_name("fire5/squeeze1x1")
118  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("fire5/relu_squeeze1x1");
119  graph << get_expand_fire_node(data_path, "fire5", weights_layout, 128U, 128U).set_name("fire5/concat");
120  graph << ConvolutionLayer(
121  1U, 1U, 48U,
122  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire6_squeeze1x1_w.npy", weights_layout),
123  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire6_squeeze1x1_b.npy"),
124  PadStrideInfo(1, 1, 0, 0))
125  .set_name("fire6/squeeze1x1")
126  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("fire6/relu_squeeze1x1");
127  graph << get_expand_fire_node(data_path, "fire6", weights_layout, 192U, 192U).set_name("fire6/concat");
128  graph << ConvolutionLayer(
129  1U, 1U, 48U,
130  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire7_squeeze1x1_w.npy", weights_layout),
131  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire7_squeeze1x1_b.npy"),
132  PadStrideInfo(1, 1, 0, 0))
133  .set_name("fire7/squeeze1x1")
134  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("fire7/relu_squeeze1x1");
135  graph << get_expand_fire_node(data_path, "fire7", weights_layout, 192U, 192U).set_name("fire7/concat");
136  graph << ConvolutionLayer(
137  1U, 1U, 64U,
138  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire8_squeeze1x1_w.npy", weights_layout),
139  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire8_squeeze1x1_b.npy"),
140  PadStrideInfo(1, 1, 0, 0))
141  .set_name("fire8/squeeze1x1")
142  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("fire8/relu_squeeze1x1");
143  graph << get_expand_fire_node(data_path, "fire8", weights_layout, 256U, 256U).set_name("fire8/concat");
144  graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool8")
145  << ConvolutionLayer(
146  1U, 1U, 64U,
147  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire9_squeeze1x1_w.npy", weights_layout),
148  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/fire9_squeeze1x1_b.npy"),
149  PadStrideInfo(1, 1, 0, 0))
150  .set_name("fire9/squeeze1x1")
151  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("fire9/relu_squeeze1x1");
152  graph << get_expand_fire_node(data_path, "fire9", weights_layout, 256U, 256U).set_name("fire9/concat");
153  graph << ConvolutionLayer(
154  1U, 1U, 1000U,
155  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv10_w.npy", weights_layout),
156  get_weights_accessor(data_path, "/cnn_data/squeezenet_v1.0_model/conv10_b.npy"),
157  PadStrideInfo(1, 1, 0, 0))
158  .set_name("conv10")
159  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("relu_conv10")
160  << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, operation_layout)).set_name("pool10")
161  << FlattenLayer().set_name("flatten")
162  << SoftmaxLayer().set_name("prob")
163  << OutputLayer(get_output_accessor(common_params, 5));
164 
165  // Finalize graph
166  GraphConfig config;
167  config.num_threads = common_params.threads;
168  config.use_tuner = common_params.enable_tuner;
169  config.tuner_mode = common_params.tuner_mode;
170  config.tuner_file = common_params.tuner_file;
171  config.mlgo_file = common_params.mlgo_file;
172  config.convert_to_uint8 = (common_params.data_type == DataType::QASYMM8);
173 
174  graph.finalize(common_params.target, config);
175 
176  return true;
177  }
178  void do_run() override
179  {
180  // Run graph
181  graph.run();
182  }
183 
184 private:
185  CommandLineParser cmd_parser;
186  CommonGraphOptions common_opts;
187  CommonGraphParams common_params;
188  Stream graph;
189 
190  ConcatLayer get_expand_fire_node(const std::string &data_path, std::string &&param_path, DataLayout weights_layout,
191  unsigned int expand1_filt, unsigned int expand3_filt)
192  {
193  std::string total_path = "/cnn_data/squeezenet_v1.0_model/" + param_path + "_";
194  SubStream i_a(graph);
195  i_a << ConvolutionLayer(
196  1U, 1U, expand1_filt,
197  get_weights_accessor(data_path, total_path + "expand1x1_w.npy", weights_layout),
198  get_weights_accessor(data_path, total_path + "expand1x1_b.npy"),
199  PadStrideInfo(1, 1, 0, 0))
200  .set_name(param_path + "/expand1x1")
201  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_expand1x1");
202 
203  SubStream i_b(graph);
204  i_b << ConvolutionLayer(
205  3U, 3U, expand3_filt,
206  get_weights_accessor(data_path, total_path + "expand3x3_w.npy", weights_layout),
207  get_weights_accessor(data_path, total_path + "expand3x3_b.npy"),
208  PadStrideInfo(1, 1, 1, 1))
209  .set_name(param_path + "/expand3x3")
210  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_expand3x3");
211 
212  return ConcatLayer(std::move(i_a), std::move(i_b));
213  }
214 };
215 
216 /** Main program for Squeezenet v1.0
217  *
218  * Model is based on:
219  * https://arxiv.org/abs/1602.07360
220  * "SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size"
221  * Forrest N. Iandola, Song Han, Matthew W. Moskewicz, Khalid Ashraf, William J. Dally, Kurt Keutzer
222  *
223  * Provenance: https://github.com/DeepScale/SqueezeNet/blob/master/SqueezeNet_v1.0/squeezenet_v1.0.caffemodel
224  *
225  * @note To list all the possible arguments execute the binary appended with the --help option
226  *
227  * @param[in] argc Number of arguments
228  * @param[in] argv Arguments
229  */
230 int main(int argc, char **argv)
231 {
232  return arm_compute::utils::run_example<GraphSqueezenetExample>(argc, argv);
233 }
Graph configuration structure Device target types.
Definition: Types.h:80
CLTunerMode tuner_mode
Tuner mode to be used by the CL tuner.
Definition: Types.h:87
std::unique_ptr< graph::ITensorAccessor > get_input_accessor(const arm_compute::utils::CommonGraphParams &graph_parameters, std::unique_ptr< IPreprocessor > preprocessor=nullptr, bool bgr=true)
Generates appropriate input accessor according to the specified graph parameters. ...
Definition: GraphUtils.h:497
int main(int argc, char **argv)
Main program for Squeezenet v1.0.
bool convert_to_uint8
Convert graph to a synthetic uint8 graph.
Definition: Types.h:86
void consume_common_graph_parameters(CommonGraphValidateOptions &options, CommonParams &common_params)
Consumes the consume_common_graph_parameters graph options and creates a structure containing any inf...
Includes all the Graph headers at once.
Common command line options used to configure the graph examples.
Class to parse command line arguments.
std::string mlgo_file
Filename to load MLGO heuristics from.
Definition: Types.h:90
std::string tuner_file
File to load/store tuning values from.
Definition: Types.h:89
quantized, asymmetric fixed-point 8-bit number unsigned
Abstract Example class.
Definition: Utils.h:78
Num samples, channels, height, width.
TensorShape permute_shape(TensorShape tensor_shape, DataLayout in_data_layout, DataLayout out_data_layout)
Permutes a given tensor shape given the input and output data layout.
Definition: GraphUtils.h:664
TensorDescriptor & set_layout(DataLayout data_layout)
Sets tensor descriptor data layout.
Structure holding all the common graph parameters.
std::unique_ptr< graph::ITensorAccessor > get_output_accessor(const arm_compute::utils::CommonGraphParams &graph_parameters, size_t top_n=5, bool is_validation=false, std::ostream &output_stream=std::cout)
Generates appropriate output accessor according to the specified graph parameters.
Definition: GraphUtils.h:543
bool use_tuner
Use a tuner in tunable backends.
Definition: Types.h:85
std::unique_ptr< graph::ITensorAccessor > get_weights_accessor(const std::string &path, const std::string &data_file, DataLayout file_layout=DataLayout::NCHW)
Generates appropriate weights accessor according to the specified path.
Definition: GraphUtils.h:475
int num_threads
Number of threads to use (thread capable backends), if 0 the backend will auto-initialize, if -1 the backend will stay as it is.
Definition: Types.h:88
Stream frontend class to construct simple graphs in a stream fashion.
Definition: Stream.h:45
DataLayout
[DataLayout enum definition]
Definition: Types.h:120
ILayer & set_name(std::string name)
Sets the name of the layer.
Definition: ILayer.h:55