1 //
2 // Copyright © 2017 Arm Ltd. All rights reserved.
3 // SPDX-License-Identifier: MIT
4 //
5 #include "../InferenceTest.hpp"
6 #include "../ImagePreprocessor.hpp"
7 #include "armnnTfLiteParser/ITfLiteParser.hpp"
8 
9 #include <cxxopts/cxxopts.hpp>
10 #include <fstream>
11 
12 using namespace armnnTfLiteParser;
13 
ParseDataset(const std::string & filename)14 std::vector<ImageSet> ParseDataset(const std::string& filename)
15 {
16     std::ifstream read(filename);
17     std::vector<ImageSet> imageSet;
18     if (read.is_open())
19     {
20         // Get the images and the correct corresponding label from the given file
21         for (std::string line; std::getline(read, line);)
22         {
23             stringstream ss(line);
24             std::string image_name;
25             std::string label;
26             getline(ss, image_name, ' ');
27             getline(ss, label, ' ');
28             imageSet.push_back(ImageSet(image_name, std::stoi(label)));
29         }
30     }
31     else
32     {
33         // Use the default images
34         imageSet.push_back(ImageSet("Dog.jpg", 209));
35         // top five predictions in tensorflow:
36         // -----------------------------------
37         // 209:Labrador retriever 0.949995
38         // 160:Rhodesian ridgeback 0.0270182
39         // 208:golden retriever 0.0192866
40         // 853:tennis ball 0.000470382
41         // 239:Greater Swiss Mountain dog 0.000464451
42         imageSet.push_back(ImageSet("Cat.jpg", 283));
43         // top five predictions in tensorflow:
44         // -----------------------------------
45         // 283:tiger cat 0.579016
46         // 286:Egyptian cat 0.319676
47         // 282:tabby, tabby cat 0.0873346
48         // 288:lynx, catamount 0.011163
49         // 289:leopard, Panthera pardus 0.000856755
50         imageSet.push_back(ImageSet("shark.jpg", 3));
51         // top five predictions in tensorflow:
52         // -----------------------------------
53         // 3:great white shark, white shark, ... 0.996926
54         // 4:tiger shark, Galeocerdo cuvieri 0.00270528
55         // 149:killer whale, killer, orca, ... 0.000121848
56         // 395:sturgeon 7.78977e-05
57         // 5:hammerhead, hammerhead shark 6.44127e-055
58     };
59     return imageSet;
60 }
61 
GetLabelsFilenameFromOptions(int argc,char * argv[])62 std::string GetLabelsFilenameFromOptions(int argc, char* argv[])
63 {
64     cxxopts::Options options("TfLiteMobilenetQuantized-Armnn","Validation Options");
65 
66     std::string fileName;
67     try
68     {
69         options
70             .allow_unrecognised_options()
71             .add_options()
72                 ("l,labels",
73                     "Filename of a text file where in each line contains an image "
74                     "filename and the correct label the network should predict when fed that image",
75                     cxxopts::value<std::string>(fileName));
76 
77         auto result = options.parse(argc, argv);
78     }
79     catch (const cxxopts::OptionException& e)
80     {
81         std::cerr << e.what() << std::endl;
82         exit(EXIT_FAILURE);
83     }
84     catch (const std::exception& e)
85     {
86         std::cerr << "Fatal internal error: [" << e.what() << "]" << std::endl;
87         exit(EXIT_FAILURE);
88     }
89 
90     return fileName;
91 }
92 
93 
main(int argc,char * argv[])94 int main(int argc, char* argv[])
95 {
96     int retVal = EXIT_FAILURE;
97     try
98     {
99         // Coverity fix: The following code may throw an exception of type std::length_error.
100         const std::string labels_file = GetLabelsFilenameFromOptions(argc,argv);
101         std::vector<ImageSet> imageSet = ParseDataset(labels_file);
102         std::vector<unsigned int> indices(imageSet.size());
103         std::generate(indices.begin(), indices.end(), [n = 0] () mutable { return n++; });
104 
105         armnn::TensorShape inputTensorShape({ 1, 224, 224, 3  });
106 
107         using DataType = uint8_t;
108         using DatabaseType = ImagePreprocessor<DataType>;
109         using ParserType = armnnTfLiteParser::ITfLiteParser;
110         using ModelType = InferenceModel<ParserType, DataType>;
111 
112         // Coverity fix: ClassifierInferenceTestMain() may throw uncaught exceptions.
113         retVal = armnn::test::ClassifierInferenceTestMain<DatabaseType,
114                                                           ParserType>(
115                      argc, argv,
116                      "mobilenet_v1_1.0_224_quant.tflite", // model name
117                      true,                                // model is binary
118                      "input",                             // input tensor name
119                      "MobilenetV1/Predictions/Reshape_1", // output tensor name
120                      indices,                             // vector of indices to select which images to validate
121                      [&imageSet](const char* dataDir, const ModelType &) {
122                          // we need to get the input quantization parameters from
123                          // the parsed model
124                          return DatabaseType(
125                              dataDir,
126                              224,
127                              224,
128                              imageSet,
129                              1);
130                      },
131                      &inputTensorShape);
132     }
133     catch (const std::exception& e)
134     {
135         // Coverity fix: BOOST_LOG_TRIVIAL (typically used to report errors) may throw an
136         // exception of type std::length_error.
137         // Using stderr instead in this context as there is no point in nesting try-catch blocks here.
138         std::cerr << "WARNING: " << *argv << ": An error has occurred when running "
139                      "the classifier inference tests: " << e.what() << std::endl;
140     }
141     return retVal;
142 }
143