MNIST数据库介绍及转换

MNIST数据库介绍:MNIST是一个手写数字数据库,它有60000个训练样本集和10000个测试样本集。它是NIST数据库的一个子集。

MNIST数据库官方网址为: ,也可以在windows下直接下载,train-images-idx3-ubyte.gz、train-labels-idx1-ubyte.gz等。下载四个文件,解压缩。解压缩后发现这些文件并不是标准的图像格式。这些图像数据都保存在二进制文件中。每个样本图像的宽高为28*28。

以下为将其转换成普通的jpg图像格式的代码:

#include <iostream>
#include <fstream>

#include "opencv2/core/core.hpp"
#include "opencv2/highgui/highgui.hpp"
#include "opencv2/imgproc/imgproc.hpp"

using namespace std;

int ReverseInt(int i)
{
 unsigned char ch1, ch2, ch3, ch4;
 ch1 = i & 255;
 ch2 = (i >> 8) & 255;
 ch3 = (i >> 16) & 255;
 ch4 = (i >> 24) & 255;
 return((int) ch1 << 24) + ((int)ch2 << 16) + ((int)ch3 << 8) + ch4;
}

void read_Mnist(string filename, vector<cv::Mat> &vec)
{
 ifstream file (filename, ios::binary);
 if (file.is_open()) {
  int magic_number = 0;
  int number_of_images = 0;
  int n_rows = 0;
  int n_cols = 0;
  file.read((char*) &magic_number, sizeof(magic_number));
  magic_number = ReverseInt(magic_number);
  file.read((char*) &number_of_images,sizeof(number_of_images));
  number_of_images = ReverseInt(number_of_images);
  file.read((char*) &n_rows, sizeof(n_rows));
  n_rows = ReverseInt(n_rows);
  file.read((char*) &n_cols, sizeof(n_cols));
  n_cols = ReverseInt(n_cols);

for(int i = 0; i < number_of_images; ++i) {
   cv::Mat tp = cv::Mat::zeros(n_rows, n_cols, CV_8UC1);
   for(int r = 0; r < n_rows; ++r) {
    for(int c = 0; c < n_cols; ++c) {
     unsigned char temp = 0;
     file.read((char*) &temp, sizeof(temp));
     tp.at<uchar>(r, c) = (int) temp;
    }
   }
   vec.push_back(tp);
  }
 }
}

void read_Mnist_Label(string filename, vector<int> &vec)
{
 ifstream file (filename, ios::binary);
 if (file.is_open()) {
  int magic_number = 0;
  int number_of_images = 0;
  int n_rows = 0;
  int n_cols = 0;
  file.read((char*) &magic_number, sizeof(magic_number));
  magic_number = ReverseInt(magic_number);
  file.read((char*) &number_of_images,sizeof(number_of_images));
  number_of_images = ReverseInt(number_of_images);

for(int i = 0; i < number_of_images; ++i) {
   unsigned char temp = 0;
   file.read((char*) &temp, sizeof(temp));
   vec[i]= (int)temp;
  }
 }
}

string GetImageName(int number, int arr[])
{
 string str1, str2;

for (int i = 0; i < 10; i++) {
  if (number == i) {
   arr[i]++;
   char ch1[10]; 
   sprintf(ch1, "%d", arr[i]); 
   str1 = std::string(ch1);

if (arr[i] < 10) {
    str1 = "0000" + str1;
   } else if (arr[i] < 100) {
    str1 = "000" + str1;
   } else if (arr[i] < 1000) {
    str1 = "00" + str1;
   } else if (arr[i] < 10000) {
    str1 = "0" + str1;
   }

break;
  }
 }

char ch2[10];
 sprintf(ch2, "%d", number);
 str2 = std::string(ch2);

str2 = str2 + "_" + str1;

return str2;
}

int main()
{
 //reference:
 //test images and test labels
 //read MNIST image into OpenCV Mat vector
 string filename_test_images = "D:/Download/t10k-images-idx3-ubyte/t10k-images.idx3-ubyte";
 int number_of_test_images = 10000;
    vector<cv::Mat> vec_test_images;

read_Mnist(filename_test_images, vec_test_images);

//read MNIST label into int vector
    string filename_test_labels = "D:/Download/t10k-labels-idx1-ubyte/t10k-labels.idx1-ubyte";
    vector<int> vec_test_labels(number_of_test_images);

read_Mnist_Label(filename_test_labels, vec_test_labels);

if (vec_test_images.size() != vec_test_labels.size()) {
  cout<<"parse MNIST test file error"<<endl;
  return -1;
 }

//save test images
 int count_digits[10];
 for (int i = 0; i < 10; i++)
  count_digits[i] = 0;

string save_test_images_path = "D:/Download/MNIST/test_images/";

内容版权声明:除非注明,否则皆为本站原创文章。

转载注明出处:https://www.heiqu.com/20ff44dfbb425d3dba9a6c6120670d0a.html