MNIST数据集转换为图像

     很多的深度学习框架都有以MNIST为数据集的demo，MNIST是很好的手写数字数据集。在网上很容易找到资源，但是下载下来的文件并不是普通的图片格式。不转换为图片格式也可以用。但有时，我们希望得到可视化的图片格式。

MNIST数据集包含4个文件：
train-images-idx3-ubyte:training set images 
train-labels-idx1-ubyte:training set labels 
t10k-images-idx3-ubyte: test set images 
t10k-labels-idx1-ubyte: test set labels
文件的格式很简单，可以理解为一个很长的一维数组。
测试图像(rain-images-idx3-ubyte)与训练图像(train-images-idx3-ubyte)由5部分组成：

32bits int (magic number)

32bits int 图像个数

32bits int 图像高度28

32bits int 图像宽度28

像素值 （pixels）

 
测试标签(t10k-labels-idx1-ubyte)与训练标签(train-labels-idx1-ubyte)由3部分组成：

32bits int (magic number)

32bits int 图像个数

标签 （labels）

知道了文件的格式，写一个简单的程序就可以把MNIST数据集转换为图像。

<span style="font-family:SimSun;">#include <iostream>
#include <fstream>

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

using namespace std;

//英特尔处理器和其他低端机用户必须翻转头字节。
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;
}
//读取训练与测试数据
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;
//从文件中读取sizeof(magic_number) 个字符到 &magic_number
file.read((char*) &magic_number, sizeof(magic_number));
magic_number = ReverseInt(magic_number);
//获取训练或测试image的个数number_of_images
file.read((char*) &number_of_images,sizeof(number_of_images));
number_of_images = ReverseInt(number_of_images);
//获取训练或测试图像的高度Heigh
file.read((char*) &n_rows, sizeof(n_rows));
n_rows = ReverseInt(n_rows);
//获取训练或测试图像的宽度Width
file.read((char*) &n_cols, sizeof(n_cols));
n_cols = ReverseInt(n_cols);
//获取第i幅图像，保存到vec中
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);
}
}
}

//读取训练与测试标签
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()
{
//测试数据和测试标签
//读取测试数据 转换为Mat
string filename_test_images = "D:/Mycode/t10k-images-idx3-ubyte/t10k-images.idx3-ubyte";
int number_of_test_images = 10000;   //测试数据10000个
vector<cv::Mat> vec_test_images;

read_Mnist(filename_test_images, vec_test_images);

//读取测试标签 转换为vector
string filename_test_labels = "D:/Mycode/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;
}

//保存测试图像
int count_digits[10];
for (int i = 0; i < 10; i++)
count_digits[i] = 0;

string save_test_images_path = "D:/Mycode/MNIST/test_images/";   //保存路径

for (int i = 0; i < vec_test_images.size(); i++)
{
int number = vec_test_labels[i];
string image_name = GetImageName(number, count_digits);
image_name = save_test_images_path + image_name + ".jpg";

cv::imwrite(image_name, vec_test_images[i]);
}

//训练数据与训练标签
//read MNIST image into OpenCV Mat vector
string filename_train_images = "D:/Mycode/train-images-idx3-ubyte/train-images.idx3-ubyte";
int number_of_train_images = 60000;
vector<cv::Mat> vec_train_images;

read_Mnist(filename_train_images, vec_train_images);

//read MNIST label into int vector
string filename_train_labels = "D:/Mycode/train-labels-idx1-ubyte/train-labels.idx1-ubyte";
vector<int> vec_train_labels(number_of_train_images);

read_Mnist_Label(filename_train_labels, vec_train_labels);

if (vec_train_images.size() != vec_train_labels.size()) {
cout<<"parse MNIST train file error"<<endl;
return -1;
}

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

string save_train_images_path = "D:/Mycode/MNIST/train_images/"; //保存路径

for (int i = 0; i < vec_train_images.size(); i++) {
int number = vec_train_labels[i];
string image_name = GetImageName(number, count_digits);
image_name = save_train_images_path + image_name + ".jpg";

cv::imwrite(image_name, vec_train_images[i]);
}

return 1;
}</span>

程序来源：http://eric-yuan.me/cpp-read-mnist/

参考：http://yann.lecun.com/exdb/mnist/      http://blog.csdn.net/fengbingchun/article/details/49611549