Compute Library
 21.02
graph_googlenet.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 Googlenet's network using the Compute Library's graph API */
35 class GraphGooglenetExample : public Example
36 {
37 public:
38  GraphGooglenetExample()
39  : cmd_parser(), common_opts(cmd_parser), common_params(), graph(0, "GoogleNet")
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  // Checks
59  ARM_COMPUTE_EXIT_ON_MSG(arm_compute::is_data_type_quantized_asymmetric(common_params.data_type), "QASYMM8 not supported for this graph");
60 
61  // Print parameter values
62  std::cout << common_params << std::endl;
63 
64  // Get trainable parameters data path
65  std::string data_path = common_params.data_path;
66 
67  // Create a preprocessor object
68  const std::array<float, 3> mean_rgb{ { 122.68f, 116.67f, 104.01f } };
69  std::unique_ptr<IPreprocessor> preprocessor = std::make_unique<CaffePreproccessor>(mean_rgb);
70 
71  // Create input descriptor
72  const auto operation_layout = common_params.data_layout;
73  const TensorShape tensor_shape = permute_shape(TensorShape(224U, 224U, 3U, 1U), DataLayout::NCHW, operation_layout);
74  TensorDescriptor input_descriptor = TensorDescriptor(tensor_shape, common_params.data_type).set_layout(operation_layout);
75 
76  // Set weights trained layout
77  const DataLayout weights_layout = DataLayout::NCHW;
78 
79  graph << common_params.target
80  << common_params.fast_math_hint
81  << InputLayer(input_descriptor, get_input_accessor(common_params, std::move(preprocessor)))
83  7U, 7U, 64U,
84  get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv1/conv1_7x7_s2_w.npy", weights_layout),
85  get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv1/conv1_7x7_s2_b.npy"),
86  PadStrideInfo(2, 2, 3, 3))
87  .set_name("conv1/7x7_s2")
88  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv1/relu_7x7")
89  << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool1/3x3_s2")
90  << NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f)).set_name("pool1/norm1")
92  1U, 1U, 64U,
93  get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_reduce_w.npy", weights_layout),
94  get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_reduce_b.npy"),
95  PadStrideInfo(1, 1, 0, 0))
96  .set_name("conv2/3x3_reduce")
97  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv2/relu_3x3_reduce")
99  3U, 3U, 192U,
100  get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_w.npy", weights_layout),
101  get_weights_accessor(data_path, "/cnn_data/googlenet_model/conv2/conv2_3x3_b.npy"),
102  PadStrideInfo(1, 1, 1, 1))
103  .set_name("conv2/3x3")
104  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name("conv2/relu_3x3")
105  << NormalizationLayer(NormalizationLayerInfo(NormType::CROSS_MAP, 5, 0.0001f, 0.75f)).set_name("conv2/norm2")
106  << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool2/3x3_s2");
107  graph << get_inception_node(data_path, "inception_3a", weights_layout, 64, std::make_tuple(96U, 128U), std::make_tuple(16U, 32U), 32U).set_name("inception_3a/concat");
108  graph << get_inception_node(data_path, "inception_3b", weights_layout, 128, std::make_tuple(128U, 192U), std::make_tuple(32U, 96U), 64U).set_name("inception_3b/concat");
109  graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool3/3x3_s2");
110  graph << get_inception_node(data_path, "inception_4a", weights_layout, 192, std::make_tuple(96U, 208U), std::make_tuple(16U, 48U), 64U).set_name("inception_4a/concat");
111  graph << get_inception_node(data_path, "inception_4b", weights_layout, 160, std::make_tuple(112U, 224U), std::make_tuple(24U, 64U), 64U).set_name("inception_4b/concat");
112  graph << get_inception_node(data_path, "inception_4c", weights_layout, 128, std::make_tuple(128U, 256U), std::make_tuple(24U, 64U), 64U).set_name("inception_4c/concat");
113  graph << get_inception_node(data_path, "inception_4d", weights_layout, 112, std::make_tuple(144U, 288U), std::make_tuple(32U, 64U), 64U).set_name("inception_4d/concat");
114  graph << get_inception_node(data_path, "inception_4e", weights_layout, 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U).set_name("inception_4e/concat");
115  graph << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, operation_layout, PadStrideInfo(2, 2, 0, 0, DimensionRoundingType::CEIL))).set_name("pool4/3x3_s2");
116  graph << get_inception_node(data_path, "inception_5a", weights_layout, 256, std::make_tuple(160U, 320U), std::make_tuple(32U, 128U), 128U).set_name("inception_5a/concat");
117  graph << get_inception_node(data_path, "inception_5b", weights_layout, 384, std::make_tuple(192U, 384U), std::make_tuple(48U, 128U), 128U).set_name("inception_5b/concat");
118  graph << PoolingLayer(PoolingLayerInfo(PoolingType::AVG, 7, operation_layout, PadStrideInfo(1, 1, 0, 0, DimensionRoundingType::CEIL))).set_name("pool5/7x7_s1")
120  1000U,
121  get_weights_accessor(data_path, "/cnn_data/googlenet_model/loss3/loss3_classifier_w.npy", weights_layout),
122  get_weights_accessor(data_path, "/cnn_data/googlenet_model/loss3/loss3_classifier_b.npy"))
123  .set_name("loss3/classifier")
124  << SoftmaxLayer().set_name("prob")
125  << OutputLayer(get_output_accessor(common_params, 5));
126 
127  // Finalize graph
128  GraphConfig config;
129  config.num_threads = common_params.threads;
130  config.use_tuner = common_params.enable_tuner;
131  config.tuner_mode = common_params.tuner_mode;
132  config.tuner_file = common_params.tuner_file;
133  config.mlgo_file = common_params.mlgo_file;
134 
135  graph.finalize(common_params.target, config);
136 
137  return true;
138  }
139  void do_run() override
140  {
141  // Run graph
142  graph.run();
143  }
144 
145 private:
146  CommandLineParser cmd_parser;
147  CommonGraphOptions common_opts;
148  CommonGraphParams common_params;
149  Stream graph;
150 
151  ConcatLayer get_inception_node(const std::string &data_path, std::string &&param_path, DataLayout weights_layout,
152  unsigned int a_filt,
153  std::tuple<unsigned int, unsigned int> b_filters,
154  std::tuple<unsigned int, unsigned int> c_filters,
155  unsigned int d_filt)
156  {
157  std::string total_path = "/cnn_data/googlenet_model/" + param_path + "/" + param_path + "_";
158  SubStream i_a(graph);
159  i_a << ConvolutionLayer(
160  1U, 1U, a_filt,
161  get_weights_accessor(data_path, total_path + "1x1_w.npy", weights_layout),
162  get_weights_accessor(data_path, total_path + "1x1_b.npy"),
163  PadStrideInfo(1, 1, 0, 0))
164  .set_name(param_path + "/1x1")
165  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_1x1");
166 
167  SubStream i_b(graph);
168  i_b << ConvolutionLayer(
169  1U, 1U, std::get<0>(b_filters),
170  get_weights_accessor(data_path, total_path + "3x3_reduce_w.npy", weights_layout),
171  get_weights_accessor(data_path, total_path + "3x3_reduce_b.npy"),
172  PadStrideInfo(1, 1, 0, 0))
173  .set_name(param_path + "/3x3_reduce")
174  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_3x3_reduce")
175  << ConvolutionLayer(
176  3U, 3U, std::get<1>(b_filters),
177  get_weights_accessor(data_path, total_path + "3x3_w.npy", weights_layout),
178  get_weights_accessor(data_path, total_path + "3x3_b.npy"),
179  PadStrideInfo(1, 1, 1, 1))
180  .set_name(param_path + "/3x3")
181  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_3x3");
182 
183  SubStream i_c(graph);
184  i_c << ConvolutionLayer(
185  1U, 1U, std::get<0>(c_filters),
186  get_weights_accessor(data_path, total_path + "5x5_reduce_w.npy", weights_layout),
187  get_weights_accessor(data_path, total_path + "5x5_reduce_b.npy"),
188  PadStrideInfo(1, 1, 0, 0))
189  .set_name(param_path + "/5x5_reduce")
190  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_5x5_reduce")
191  << ConvolutionLayer(
192  5U, 5U, std::get<1>(c_filters),
193  get_weights_accessor(data_path, total_path + "5x5_w.npy", weights_layout),
194  get_weights_accessor(data_path, total_path + "5x5_b.npy"),
195  PadStrideInfo(1, 1, 2, 2))
196  .set_name(param_path + "/5x5")
197  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_5x5");
198 
199  SubStream i_d(graph);
200  i_d << PoolingLayer(PoolingLayerInfo(PoolingType::MAX, 3, common_params.data_layout, PadStrideInfo(1, 1, 1, 1, DimensionRoundingType::CEIL))).set_name(param_path + "/pool")
201  << ConvolutionLayer(
202  1U, 1U, d_filt,
203  get_weights_accessor(data_path, total_path + "pool_proj_w.npy", weights_layout),
204  get_weights_accessor(data_path, total_path + "pool_proj_b.npy"),
205  PadStrideInfo(1, 1, 0, 0))
206  .set_name(param_path + "/pool_proj")
207  << ActivationLayer(ActivationLayerInfo(ActivationLayerInfo::ActivationFunction::RELU)).set_name(param_path + "/relu_pool_proj");
208 
209  return ConcatLayer(std::move(i_a), std::move(i_b), std::move(i_c), std::move(i_d));
210  }
211 };
212 
213 /** Main program for Googlenet
214  *
215  * Model is based on:
216  * https://arxiv.org/abs/1409.4842
217  * "Going deeper with convolutions"
218  * Christian Szegedy, Wei Liu, Yangqing Jia, Pierre Sermanet, Scott Reed, Dragomir Anguelov, Dumitru Erhan, Vincent Vanhoucke, Andrew Rabinovich
219  *
220  * Provenance: https://github.com/BVLC/caffe/tree/master/models/bvlc_googlenet
221  *
222  * @note To list all the possible arguments execute the binary appended with the --help option
223  *
224  * @param[in] argc Number of arguments
225  * @param[in] argv Arguments
226  */
227 int main(int argc, char **argv)
228 {
229  return arm_compute::utils::run_example<GraphGooglenetExample>(argc, argv);
230 }
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 Googlenet.
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
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
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.
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
Normalization applied cross maps.
ILayer & set_name(std::string name)
Sets the name of the layer.
Definition: ILayer.h:55