Opencv 生成训练集和测试集---CSV文件读写

Opencv 的训练集和测试写入通过format(Mat , Formatter::FMT_CSV)，方式完成；

读取是通过TrainData::loadFromCSV完成。

具体代码实现如下：

/*  * filestorage_sample demonstrate the usage of the opencv serialization functionality  */   #include "opencv2/core.hpp" #include <opencv2/imgproc.hpp> #include "opencv2/imgcodecs.hpp" #include <opencv2/highgui.hpp> #include <opencv2/ml.hpp> #include <iostream> #include <string> #include <fstream>   using std::string; using std::cout; using std::endl; using std::cerr; using std::ostream; using namespace cv; using namespace cv::ml; using namespace std;   #define IMG_WIDTH 128 #define IMG_HEIGHT 128 #define IMG_CHANNEL 1   int main(int ac, char** av) {   string filename ;   string path;   if(ac > 2 )   {     filename = av[2];     path = av[1];   }   else   {       printf(" should be ./*.out path.txt outputfilename\n");   }     printf("%s(%d):%s\n", __FUNCTION__, __LINE__, filename.c_str());     if (filename.empty())   {       printf("Error:fileName NULL\n");     return 1;   }     //write   {       ifstream fin;     string tmp;     fin.open(path.c_str()); //  FILE* fpt = fopen(filename.c_str(),(const char*)"w"); ofstream file(filename.c_str());       cout << "writing images\n";     int i = 0;     vector<string> FileVec;       while(fin >> tmp)     { #if 0         stringstream tmpStr;         tmpStr<<"image"<<i;         cout  <<tmpStr.str()<<endl;         i++; #endif         FileVec.push_back(tmp);     }       vector<string>::iterator strIt;     vector<Mat> AllImage;     Mat MergeOut;     int Cnt = 0;         for(strIt = FileVec.begin();strIt != FileVec.end(); strIt++, Cnt++ )     {         Mat inputMat = imread(*strIt, 0);         int rows = inputMat.rows;         int cols = inputMat.cols;         int channels = inputMat.channels();           vector<Mat> imgVect;         imgVect.resize(3);           printf("%d,%d,%d\n", inputMat.rows, inputMat.cols, inputMat.channels());           if(rows != IMG_WIDTH || cols != IMG_HEIGHT || channels != IMG_CHANNEL )         {             printf("Error %s: rows should be %d, cols should be %d, channels should be %d\n", strIt->c_str(), IMG_WIDTH, IMG_HEIGHT, IMG_CHANNEL);             return 0;         } #if 0         split(inputMat, imgVect);         unsigned char *pData = inputMat.data;                  ofstream file("test.csv");         file << format(imgVect[0], Formatter::FMT_CSV);         file.close(); #endif         Mat imgResize;         resize(inputMat, imgResize, Size(64,64), 64, 64, INTER_LINEAR);          // Mat ImgForTrain(1, inputMat.rows*inputMat.cols,  CV_8UC1, (void*)inputMat.data);         Mat ImgForTrain(1, imgResize.rows*imgResize.cols,  CV_8UC1, (void*)imgResize.data);         AllImage.push_back(ImgForTrain);         if(Cnt == 0)           MergeOut = ImgForTrain;         else             vconcat(ImgForTrain, MergeOut, MergeOut);         //file << format(ImgForTrain, Formatter::FMT_CSV);         printf("%d,%d,%d\n", MergeOut.rows, MergeOut.cols, MergeOut.channels());       }    // cout <<MergeOut<< endl;       printf("%d,%d,%d\n", MergeOut.rows, MergeOut.cols, MergeOut.channels());       Mat ImgLabel(MergeOut.rows, 1, CV_8UC1);     for(int i = 0; i < MergeOut.rows; i++)     {         if(i < MergeOut.rows/2)         {             ImgLabel.at<unsigned char>(i, 0) = 1;         }         else             ImgLabel.at<unsigned char>(i, 0) = 0;     }     hconcat( MergeOut, ImgLabel,  MergeOut);       file << format(MergeOut, Formatter::FMT_CSV);       //    fclose(fpt);       fin.close();     file.close();         }     //read   {     cout <<"\n"<<filename<<endl;       Ptr<TrainData> train_data;     train_data = TrainData::loadFromCSV(filename.c_str(), 0);     Mat m = train_data->getTrainSamples();     //cout << format(m, Formatter::FMT_CSV) << endl;     printf("\n%d,%d,%d\n", m.rows, m.cols, m.channels()); //    cout <<m<< endl;   //    cout << m.row(0) << endl;       Mat label = train_data->getTrainResponses();     printf("\n%d,%d,%d\n", label.rows, label.cols, label.channels());     }       return 0; }

 

该代码已验证。

通过它，我生成了用于64x64分辨率的500张戴眼镜和500张不戴眼镜的训练集和100张戴眼镜和100张不戴眼镜的验证集。

其中，第一个入参是输入的图像列表文件；第二个入参是生成的数据集的文件名。