如何根据CIFAR-10的格式制作自己的数据集（C/C++版）

首先特别感谢博主 @yhl_leo

关于CIFER-10数据集可查看官方介绍，存储信息介绍如下：

不啰嗦，直接上代码实例，图片如何存储为二进制格式的三个代码文件如下：

相应的代码及备注依次如下：

#include "BinaryDataset.h"void BinaDataset::images2BinaryFile( std::string filefolder, std::vector<std::string>& img_list, std::vector<int>& img_labels, std::string filename ){const int size_list = img_list.size();FILE *fp = fopen( filename.c_str(), "wb" );if ( fp == NULL ){std::cout << "Open error!" << std::endl;fclose(fp);return;}for ( int idx = 0; idx <size_list; ++idx ){std::string currentPath = filefolder + "\\";currentPath += img_list[idx];mat2Binary( currentPath, img_labels[idx], fp );#if 1std::cout << "image " << idx+1 << " saved." << std::endl;#endif}fclose(fp);}void BinaDataset::mat2Binary( std::string& image_file, int label, FILE*& fp ){cv::Mat image = cv::imread( image_file, IMREAD_UNCHANGED );if ( !image.data ){std::cout << "Image " << getFileName(image_file) << " load failed!"<< std::endl;}else{if ( image.channels() == 1 ){cv::cvtColor(image, image, CV_GRAY2RGB );}else{cv::cvtColor(image, image, CV_BGR2RGB );}cv::Mat image_reshaped;cv::resize( image, image_reshaped, cv::Size(_iWidth, _iHeight), CV_INTER_LINEAR );convertMat2Bin(image_reshaped, label, fp);}}/*** 将图片转化为矩阵并将每个元素写入二进制文件* @param image* @param label* @param fp*/void BinaDataset::convertMat2Bin( cv::Mat& image, int label, FILE*& fp ){fwrite(&label, sizeof(char), 1, fp);int pixelCount = image.rows * image.cols;char* pData = (char*)image.data;for ( int i=0; i<pixelCount; i++ )fwrite(&pData[i*3],   sizeof(char), 1, fp); // Rfor ( int i=0; i<pixelCount; i++ )fwrite(&pData[i*3+1],   sizeof(char), 1, fp); // Gfor ( int i=0; i<pixelCount; i++ )fwrite(&pData[i*3+2],   sizeof(char), 1, fp);  // B//delete pData;}std::string BinaDataset::getFileName( std::string & filename ){int iBeginIndex = filename.find_last_of("\\")+1;int iEndIndex   = filename.length();return filename.substr( iBeginIndex, iEndIndex - iBeginIndex );}std::vector<std::string> BinaDataset::getFileLists( std::string file_folder ){file_folder += "/*.*";const char * mystr=file_folder.c_str();std::vector<std::string> flist;std::string lineStr;std::vector<std::string> extendName;extendName.push_back("jpg");extendName.push_back("JPG");extendName.push_back("bmp");extendName.push_back("png");extendName.push_back("gif");HANDLE file;WIN32_FIND_DATA fileData;char line[1024];wchar_t fn[1000];mbstowcs( fn, mystr, 999 );file = FindFirstFile( fn, &fileData );FindNextFile( file, &fileData );while(FindNextFile( file, &fileData ))    {wcstombs( line, (const wchar_t*)fileData.cFileName, 259);lineStr = line;        // remove the files which are not imagesfor (int i = 0; i < 4; i ++){if (lineStr.find(extendName[i]) < 999){flist.push_back(lineStr);break;}}}return flist;}

#ifndef BINARY_DATASET_H#define BINARY_DATASET_H#pragma once#include <iostream>#include <vector>#include <stdio.h>#include <stdlib.h>#include <Windows.h>#include "cv.h"#include "highgui.h"using namespace std;using namespace cv;class BinaDataset{public:BinaDataset(){_iHeight = 32;_iWidth  = 32;}~CrackBinaDataset(){}public:void images2BinaryFile( std::string filefolder, std::vector<std::string>& img_list, std::vector<int>& img_labels, std::string filename );void mat2Binary( std::string& image_file, int label, FILE*& fp );void convertMat2Bin( cv::Mat& image, int label, FILE*& fp );//cv::Mat imageReshape( cv::Mat& input );std::string getFileName( std::string & filename );std::vector<std::string> getFileLists( std::string file_folder );public:int _iHeight;int _iWidth;};#endif // BINARY_DATASET_H

#include "BinaryDataset.h"void main(){std::string filefolder = "train\\"; //图片所在的文件夹BinaDataset binData;std::vector<std::string> fileLists = binData.getFileLists(filefolder); // load file nameconst int size_list = fileLists.size();std::cout << "image count: " << size_list << std::endl;std::vector<int> image_labels(size_list, 0);  // generate lables, here are all 0std::string binfile = "C:\\Samples\\train.bin";binData.images2BinaryFile( filefolder, fileLists, image_labels, binfile );}

本人转换后的结果如下：

最后，将数据放入CIFAR-10模型中，并修改一下部分参数，效果还不错！