mv_machine_learning/face_recognition/src/face_recognition_dsm.cpp

   1 /**
   2  * Copyright (c) 2022 Samsung Electronics Co., Ltd All Rights Reserved
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at
   7  *
   8  * http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <dlog.h>
  18 #include <mv_private.h>
  19 #include <algorithm>
  20
  21 #include "machine_learning_exception.h"
  22 #include "face_recognition_dsm.h"
  23 #define MAX_FEATURE_VECTOR_CNT 5
  24
  25 using namespace std;
  26 using namespace mediavision::machine_learning::exception;
  27
  28 void FaceRecognitionDSM::PrintHeader(FeaVecHeader &fvh)
  29 {
  30         LOGD("signature = %u", fvh.signature);
  31         LOGD("feature vector size = %zu", fvh.feature_size);
  32         LOGD("label count = %zu", fvh.label_cnt);
  33         LOGD("data set count = %u", fvh.data_set_cnt);
  34 }
  35
  36 FaceRecognitionDSM::FaceRecognitionDSM() : DataSetManager()
  37 {}
  38
  39 bool FaceRecognitionDSM::IsFeatureVectorAllowed(unsigned int label_idx)
  40 {
  41         return (_fv_cnt_per_label[label_idx] < MAX_FEATURE_VECTOR_CNT);
  42 }
  43
  44 void FaceRecognitionDSM::AddDataSet(std::vector<float> &feature_vec, const unsigned int label_idx,
  45                                                                         const unsigned int label_cnt)
  46 {
  47         _data.push_back(feature_vec);
  48         _label_index.push_back(label_idx);
  49         _fv_cnt_per_label[label_idx]++;
  50
  51         vector<float> oneHotEncoding;
  52
  53         for (size_t num = 0; num < label_cnt; ++num)
  54                 oneHotEncoding.push_back(label_idx == num ? 1.0f : 0.0f);
  55
  56         _labels.push_back(oneHotEncoding);
  57         _feature_vector_size = feature_vec.size();
  58         _label_count = label_cnt;
  59 }
  60
  61 void FaceRecognitionDSM::LoadDataSet(const string file_name, unsigned int new_label_cnt)
  62 {
  63         std::ifstream inFile(file_name);
  64
  65         if (!inFile.is_open())
  66                 throw InvalidOperation("fail to open a file.");
  67
  68         // Feature vector file header is written at the end of the data file
  69         // So read feature vector header from the end of the file.
  70         inFile.seekg(static_cast<int>(sizeof(FeaVecHeader)) * -1, ios::end);
  71
  72         FeaVecHeader fvh;
  73
  74         inFile.read((char *) &fvh, sizeof(FeaVecHeader));
  75         if (inFile.gcount() != sizeof(FeaVecHeader))
  76                 throw InvalidOperation("Invalid feature vector file.");
  77
  78         inFile.seekg(0, ios::beg);
  79
  80         PrintHeader(fvh);
  81
  82         if (FeatureVectorManager::feature_vector_signature != fvh.signature)
  83                 throw InvalidOperation("Wrong feature vector header.");
  84
  85         /*
  86         * stride line format is as follows
  87         * ********************************
  88         * ____________________________
  89         * |feature vector|label index|
  90         * ----------------------------
  91         */
  92         size_t line_size_in_bytes = fvh.feature_size * sizeof(float) + sizeof(unsigned int);
  93
  94         _feature_vector_size = fvh.feature_size;
  95         _label_count = fvh.label_cnt;
  96
  97         vector<float> line_data(fvh.feature_size + 1);
  98
  99         for (size_t idx = 0; idx < fvh.data_set_cnt; ++idx) {
 100                 inFile.read(reinterpret_cast<char *>(line_data.data()), line_size_in_bytes);
 101
 102                 vector<float> data;
 103
 104                 copy_n(line_data.begin(), _feature_vector_size, back_inserter(data));
 105                 _data.push_back(data);
 106
 107                 unsigned int label_idx;
 108
 109                 memcpy(&label_idx, reinterpret_cast<void *>(line_data.data() + _feature_vector_size), sizeof(unsigned int));
 110
 111                 vector<float> label;
 112
 113                 // max label count may be changed so update one hot encoding table.
 114                 for (size_t num = 0; num < new_label_cnt; ++num)
 115                         label.push_back(label_idx == num ? 1.0f : 0.0f);
 116
 117                 _labels.push_back(label);
 118                 _label_index.push_back(label_idx);
 119
 120                 _fv_cnt_per_label[label_idx]++;
 121         }
 122 }