Compute Library
 21.02
graph_vgg19.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 /** Example demonstrating how to implement VGG19's network using the Compute Library's graph API */
34 class GraphVGG19Example : public Example
35 {
36 public:
37  GraphVGG19Example()
38  : cmd_parser(), common_opts(cmd_parser), common_params(), graph(0, "VGG19")
39  {
40  }
41  bool do_setup(int argc, char **argv) override
42  {
43  // Parse arguments
44  cmd_parser.parse(argc, argv);
45  cmd_parser.validate();
46 
47  // Consume common parameters
48  common_params = consume_common_graph_parameters(common_opts);
49 
50  // Return when help menu is requested
51  if(common_params.help)
52  {
53  cmd_parser.print_help(argv[0]);
54  return false;
55  }
56 
57  // Print parameter values
58  std::cout << common_params << std::endl;
59 
60  // Get trainable parameters data path
61  std::string data_path = common_params.data_path;
62 
63  // Create a preprocessor object
64  const std::array<float, 3> mean_rgb{ { 123.68f, 116.779f, 103.939f } };
65  std::unique_ptr<IPreprocessor> preprocessor = std::make_unique<CaffePreproccessor>(mean_rgb);
66 
67  // Create input descriptor
68  const auto operation_layout = common_params.data_layout;
69  const TensorShape tensor_shape = permute_shape(TensorShape(224U, 224U, 3U, 1U), DataLayout::NCHW, operation_layout);
70  TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(operation_layout);
71 
72  // Set weights trained layout
73  const DataLayout weights_layout = DataLayout::NCHW;
74 
75  graph << common_params.target
76  << common_params.fast_math_hint
77  << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor)))
78  // Layer 1
80  3U, 3U, 64U,
81  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv1_1_w.npy", weights_layout),
82  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv1_1_b.npy"),
83  PadStrideInfo(1, 1, 1, 1))
84  .set_name("conv1_1")
85  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv1_1/Relu")
87  3U, 3U, 64U,
88  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv1_2_w.npy", weights_layout),
89  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv1_2_b.npy"),
90  PadStrideInfo(1, 1, 1, 1))
91  .set_name("conv1_2")
92  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv1_2/Relu")
93  << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, operation_layout, PadStrideInfo(2, 2, 0, 0))).set_name("pool1")
94  // Layer 2
96  3U, 3U, 128U,
97  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv2_1_w.npy", weights_layout),
98  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv2_1_b.npy"),
99  PadStrideInfo(1, 1, 1, 1))
100  .set_name("conv2_1")
101  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv2_1/Relu")
102  << ConvolutionLayer(
103  3U, 3U, 128U,
104  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv2_2_w.npy", weights_layout),
105  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv2_2_b.npy"),
106  PadStrideInfo(1, 1, 1, 1))
107  .set_name("conv2_2")
108  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv2_2/Relu")
109  << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, operation_layout, PadStrideInfo(2, 2, 0, 0))).set_name("pool2")
110  // Layer 3
111  << ConvolutionLayer(
112  3U, 3U, 256U,
113  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_1_w.npy", weights_layout),
114  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_1_b.npy"),
115  PadStrideInfo(1, 1, 1, 1))
116  .set_name("conv3_1")
117  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv3_1/Relu")
118  << ConvolutionLayer(
119  3U, 3U, 256U,
120  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_2_w.npy", weights_layout),
121  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_2_b.npy"),
122  PadStrideInfo(1, 1, 1, 1))
123  .set_name("conv3_2")
124  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv3_2/Relu")
125  << ConvolutionLayer(
126  3U, 3U, 256U,
127  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_3_w.npy", weights_layout),
128  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_3_b.npy"),
129  PadStrideInfo(1, 1, 1, 1))
130  .set_name("conv3_3")
131  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv3_3/Relu")
132  << ConvolutionLayer(
133  3U, 3U, 256U,
134  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_4_w.npy", weights_layout),
135  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv3_4_b.npy"),
136  PadStrideInfo(1, 1, 1, 1))
137  .set_name("conv3_4")
138  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv3_4/Relu")
139  << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, operation_layout, PadStrideInfo(2, 2, 0, 0))).set_name("pool3")
140  // Layer 4
141  << ConvolutionLayer(
142  3U, 3U, 512U,
143  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_1_w.npy", weights_layout),
144  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_1_b.npy"),
145  PadStrideInfo(1, 1, 1, 1))
146  .set_name("conv4_1")
147  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv4_1/Relu")
148  << ConvolutionLayer(
149  3U, 3U, 512U,
150  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_2_w.npy", weights_layout),
151  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_2_b.npy"),
152  PadStrideInfo(1, 1, 1, 1))
153  .set_name("conv4_2")
154  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv4_2/Relu")
155  << ConvolutionLayer(
156  3U, 3U, 512U,
157  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_3_w.npy", weights_layout),
158  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_3_b.npy"),
159  PadStrideInfo(1, 1, 1, 1))
160  .set_name("conv4_3")
161  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv4_3/Relu")
162  << ConvolutionLayer(
163  3U, 3U, 512U,
164  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_4_w.npy", weights_layout),
165  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv4_4_b.npy"),
166  PadStrideInfo(1, 1, 1, 1))
167  .set_name("conv4_4")
168  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv4_4/Relu")
169  << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, operation_layout, PadStrideInfo(2, 2, 0, 0))).set_name("pool4")
170  // Layer 5
171  << ConvolutionLayer(
172  3U, 3U, 512U,
173  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_1_w.npy", weights_layout),
174  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_1_b.npy"),
175  PadStrideInfo(1, 1, 1, 1))
176  .set_name("conv5_1")
177  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv5_1/Relu")
178  << ConvolutionLayer(
179  3U, 3U, 512U,
180  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_2_w.npy", weights_layout),
181  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_2_b.npy"),
182  PadStrideInfo(1, 1, 1, 1))
183  .set_name("conv5_2")
184  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv5_2/Relu")
185  << ConvolutionLayer(
186  3U, 3U, 512U,
187  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_3_w.npy", weights_layout),
188  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_3_b.npy"),
189  PadStrideInfo(1, 1, 1, 1))
190  .set_name("conv5_3")
191  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv5_3/Relu")
192  << ConvolutionLayer(
193  3U, 3U, 512U,
194  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_4_w.npy", weights_layout),
195  get_weights_accessor(data_path, "/cnn_data/vgg19_model/conv5_4_b.npy"),
196  PadStrideInfo(1, 1, 1, 1))
197  .set_name("conv5_4")
198  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv5_4/Relu")
199  << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 2, operation_layout, PadStrideInfo(2, 2, 0, 0))).set_name("pool5")
200  // Layer 6
202  4096U,
203  get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc6_w.npy", weights_layout),
204  get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc6_b.npy"))
205  .set_name("fc6")
207  // Layer 7
209  4096U,
210  get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc7_w.npy", weights_layout),
211  get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc7_b.npy"))
212  .set_name("fc7")
214  // Layer 8
216  1000U,
217  get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc8_w.npy", weights_layout),
218  get_weights_accessor(data_path, "/cnn_data/vgg19_model/fc8_b.npy"))
219  .set_name("fc8")
220  // Softmax
221  << SoftmaxLayer().set_name("prob")
222  << OutputLayer(get_output_accessor(common_params, 5));
223 
224  // Finalize graph
225  GraphConfig config;
226  config.num_threads = common_params.threads;
227  config.use_tuner = common_params.enable_tuner;
228  config.tuner_mode = common_params.tuner_mode;
229  config.tuner_file = common_params.tuner_file;
230  config.mlgo_file = common_params.mlgo_file;
231  config.convert_to_uint8 = (common_params.data_type == DataType::QASYMM8);
232 
233  graph.finalize(common_params.target, config);
234 
235  return true;
236  }
237  void do_run() override
238  {
239  // Run graph
240  graph.run();
241  }
242 
243 private:
244  CommandLineParser cmd_parser;
245  CommonGraphOptions common_opts;
246  CommonGraphParams common_params;
247  Stream graph;
248 };
249 
250 /** Main program for VGG19
251  *
252  * Model is based on:
253  * https://arxiv.org/abs/1409.1556
254  * "Very Deep Convolutional Networks for Large-Scale Image Recognition"
255  * Karen Simonyan, Andrew Zisserman
256  *
257  * Provenance: www.robots.ox.ac.uk/~vgg/software/very_deep/caffe/VGG_ILSVRC_19_layers.caffemodel
258  *
259  * @note To list all the possible arguments execute the binary appended with the --help option
260  *
261  * @param[in] argc Number of arguments
262  * @param[in] argv Arguments
263  */
264 int main(int argc, char **argv)
265 {
266  return arm_compute::utils::run_example<GraphVGG19Example>(argc, argv);
267 }
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
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
int main(int argc, char **argv)
Main program for VGG19.
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