Compute Library
 21.02
graph_resnet12.cpp
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2018-2021 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 ResNet12 network using the Compute Library's graph API */
35 class GraphResNet12Example : public Example
36 {
37 public:
38  GraphResNet12Example()
39  : cmd_parser(), common_opts(cmd_parser), model_input_width(nullptr), model_input_height(nullptr), common_params(), graph(0, "ResNet12")
40  {
41  model_input_width = cmd_parser.add_option<SimpleOption<unsigned int>>("image-width", 192);
42  model_input_height = cmd_parser.add_option<SimpleOption<unsigned int>>("image-height", 128);
43 
44  // Add model id option
45  model_input_width->set_help("Input image width.");
46  model_input_height->set_help("Input image height.");
47  }
48  GraphResNet12Example(const GraphResNet12Example &) = delete;
49  GraphResNet12Example &operator=(const GraphResNet12Example &) = delete;
50  ~GraphResNet12Example() override = default;
51  bool do_setup(int argc, char **argv) override
52  {
53  // Parse arguments
54  cmd_parser.parse(argc, argv);
55  cmd_parser.validate();
56 
57  // Consume common parameters
58  common_params = consume_common_graph_parameters(common_opts);
59 
60  // Return when help menu is requested
61  if(common_params.help)
62  {
63  cmd_parser.print_help(argv[0]);
64  return false;
65  }
66 
67  // Get input image width and height
68  const unsigned int image_width = model_input_width->value();
69  const unsigned int image_height = model_input_height->value();
70 
71  // Checks
72  ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "QASYMM8 not supported for this graph");
73 
74  // Print parameter values
75  std::cout << common_params << std::endl;
76  std::cout << "Image width: " << image_width << std::endl;
77  std::cout << "Image height: " << image_height << std::endl;
78 
79  // Get trainable parameters data path
80  const std::string data_path = common_params.data_path;
81  const std::string model_path = "/cnn_data/resnet12_model/";
82 
83  // Create a preprocessor object
84  std::unique_ptr<IPreprocessor> preprocessor = std::make_unique<TFPreproccessor>();
85 
86  // Create input descriptor
87  const TensorShape tensor_shape = permute_shape(TensorShape(image_width, image_height, 3U, 1U), DataLayout::NCHW, common_params.data_layout);
88  TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(common_params.data_layout);
89 
90  // Set weights trained layout
91  const DataLayout weights_layout = DataLayout::NCHW;
92 
93  graph << common_params.target
94  << common_params.fast_math_hint
95  << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor), false /* Do not convert to BGR */))
97  9U, 9U, 64U,
98  get_weights_accessor(data_path, "conv1_weights.npy", weights_layout),
99  get_weights_accessor(data_path, "conv1_biases.npy", weights_layout),
100  PadStrideInfo(1, 1, 4, 4))
101  .set_name("conv1/convolution")
102  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv1/Relu");
103 
104  add_residual_block(data_path, "block1", weights_layout);
105  add_residual_block(data_path, "block2", weights_layout);
106  add_residual_block(data_path, "block3", weights_layout);
107  add_residual_block(data_path, "block4", weights_layout);
108 
109  graph << ConvolutionLayer(
110  3U, 3U, 64U,
111  get_weights_accessor(data_path, "conv10_weights.npy", weights_layout),
112  get_weights_accessor(data_path, "conv10_biases.npy"),
113  PadStrideInfo(1, 1, 1, 1))
114  .set_name("conv10/convolution")
115  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv10/Relu")
116  << ConvolutionLayer(
117  3U, 3U, 64U,
118  get_weights_accessor(data_path, "conv11_weights.npy", weights_layout),
119  get_weights_accessor(data_path, "conv11_biases.npy"),
120  PadStrideInfo(1, 1, 1, 1))
121  .set_name("conv11/convolution")
122  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv11/Relu")
123  << ConvolutionLayer(
124  9U, 9U, 3U,
125  get_weights_accessor(data_path, "conv12_weights.npy", weights_layout),
126  get_weights_accessor(data_path, "conv12_biases.npy"),
127  PadStrideInfo(1, 1, 4, 4))
128  .set_name("conv12/convolution")
129  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::TANH)).set_name("conv12/Tanh")
130  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::LINEAR, 0.58f, 0.5f)).set_name("conv12/Linear")
131  << OutputLayer(std::make_unique<DummyAccessor>(0));
132 
133  // Finalize graph
134  GraphConfig config;
135  config.num_threads = common_params.threads;
136  config.use_tuner = common_params.enable_tuner;
137  config.tuner_mode = common_params.tuner_mode;
138  config.tuner_file = common_params.tuner_file;
139  config.mlgo_file = common_params.mlgo_file;
140 
141  graph.finalize(common_params.target, config);
142 
143  return true;
144  }
145 
146  void do_run() override
147  {
148  // Run graph
149  graph.run();
150  }
151 
152 private:
153  CommandLineParser cmd_parser;
154  CommonGraphOptions common_opts;
155  SimpleOption<unsigned int> *model_input_width{ nullptr };
156  SimpleOption<unsigned int> *model_input_height{ nullptr };
157  CommonGraphParams common_params;
158  Stream graph;
159 
160  void add_residual_block(const std::string &data_path, const std::string &name, DataLayout weights_layout)
161  {
162  std::stringstream unit_path_ss;
163  unit_path_ss << data_path << name << "_";
164  std::stringstream unit_name_ss;
165  unit_name_ss << name << "/";
166 
167  std::string unit_path = unit_path_ss.str();
168  std::string unit_name = unit_name_ss.str();
169 
170  SubStream left(graph);
171  SubStream right(graph);
172 
173  right << ConvolutionLayer(
174  3U, 3U, 64U,
175  get_weights_accessor(data_path, unit_path + "conv1_weights.npy", weights_layout),
176  get_weights_accessor(data_path, unit_path + "conv1_biases.npy", weights_layout),
177  PadStrideInfo(1, 1, 1, 1))
178  .set_name(unit_name + "conv1/convolution")
180  get_weights_accessor(data_path, unit_path + "conv1_BatchNorm_moving_mean.npy"),
181  get_weights_accessor(data_path, unit_path + "conv1_BatchNorm_moving_variance.npy"),
182  get_weights_accessor(data_path, unit_path + "conv1_BatchNorm_gamma.npy"),
183  get_weights_accessor(data_path, unit_path + "conv1_BatchNorm_beta.npy"),
184  0.0000100099996416f)
185  .set_name(unit_name + "conv1/BatchNorm")
186  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(unit_name + "conv1/Relu")
187 
188  << ConvolutionLayer(
189  3U, 3U, 64U,
190  get_weights_accessor(data_path, unit_path + "conv2_weights.npy", weights_layout),
191  get_weights_accessor(data_path, unit_path + "conv2_biases.npy", weights_layout),
192  PadStrideInfo(1, 1, 1, 1))
193  .set_name(unit_name + "conv2/convolution")
195  get_weights_accessor(data_path, unit_path + "conv2_BatchNorm_moving_mean.npy"),
196  get_weights_accessor(data_path, unit_path + "conv2_BatchNorm_moving_variance.npy"),
197  get_weights_accessor(data_path, unit_path + "conv2_BatchNorm_gamma.npy"),
198  get_weights_accessor(data_path, unit_path + "conv2_BatchNorm_beta.npy"),
199  0.0000100099996416f)
200  .set_name(unit_name + "conv2/BatchNorm")
201  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(unit_name + "conv2/Relu");
202 
203  graph << EltwiseLayer(std::move(left), std::move(right), EltwiseOperation::Add).set_name(unit_name + "add");
204  }
205 };
206 
207 /** Main program for ResNet12
208  *
209  * Model is based on:
210  * https://arxiv.org/pdf/1709.01118.pdf
211  * "WESPE: Weakly Supervised Photo Enhancer for Digital Cameras"
212  * Andrey Ignatov, Nikolay Kobyshev, Kenneth Vanhoey, Radu Timofte, Luc Van Gool
213  *
214  * @note To list all the possible arguments execute the binary appended with the --help option
215  *
216  * @param[in] argc Number of arguments
217  * @param[in] argv Arguments
218  */
219 int main(int argc, char **argv)
220 {
221  return arm_compute::utils::run_example<GraphResNet12Example>(argc, argv);
222 }
Graph configuration structure Device target types.
Definition: Types.h:80
int main(int argc, char **argv)
Main program for ResNet12.
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
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
#define ARM_COMPUTE_EXIT_ON_MSG(cond, msg)
If the condition is true, the given message is printed and program exits.
Definition: Error.h:379
Abstract Example class.
Definition: Utils.h:78
const T & value() const
Get the option value.
Definition: SimpleOption.h:112
Num samples, channels, height, width.
const char * name
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
bool is_data_type_quantized_asymmetric(DataType dt)
Check if a given data type is of asymmetric quantized type.
Definition: Utils.h:1190
TensorDescriptor & set_layout(DataLayout data_layout)
Sets tensor descriptor data layout.
Structure holding all the common graph parameters.
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
void set_help(std::string help)
Set the help message for the option.
Definition: Option.h:125