9/09/2015

mnist image dataset (jpg files)

The MNIST dataset is a dataset of handwritten digits, comprising 60 000 training examples and 10 000 test examples. The dataset can be downloaded from http://yann.lecun.com/exdb/mnist/.
MNIST data is provided in binary file for user convenience.

But, in my case, this binary type is more difficult.
I am needed image files.

I want to start whole process of deep learning(make leveldb file, learning, classify) using MNIST data and compare to example result in caffe model.


This site tells me a hint how to make image datas.

This is my matlab code for making jpg file.

---
%%
clc;
clear all;

% Change the filenames if you've saved the files under different names
% On some platforms, the files might be saved as
% train-images.idx3-ubyte / train-labels.idx1-ubyte


images = loadMNISTImages('t10k-images.idx3-ubyte');
labels = loadMNISTLabels('t10k-labels.idx1-ubyte');

%images = loadMNISTImages('train-images.idx3-ubyte');
%labels = loadMNISTLabels('train-labels.idx1-ubyte');

% We are using display_network from the autoencoder code
display_network(images(:,1:60000)); % Show the first 100 images
disp(labels(1:1));

% save one digit number to jpg file
%{
[w h]=size(images)
endIndex = h;

dindex=zeros(10,1);
for j=1:endIndex
aImg = reshape( images(:,j), 28, 28);
imshow(aImg);
label = labels(j);
dindex( label+1 ) = dindex( label+1 ) +1;
dataD = strcat('./data2/mnist_', num2str( label ), '_', num2str( dindex(label+1) ), '.jpg' );
%dindex
imwrite(aImg, dataD);
dataD
end

fileID = fopen('test.txt','w');
%make path and label
dindex=zeros(10,1);
for j=1:endIndex
label = labels(j);
dindex( label+1 ) = dindex( label+1 ) +1;
dataD = strcat('/data2/mnist_', num2str( label ), '_', num2str( dindex(label+1) ), '.jpg' );
fprintf(fileID,'%s %d\n',dataD, label);
end
fclose(fileID);
%}
---

And images of mnist files


If you need images of mnist, request to me, I will send by email.
The file size is approximately 37 MB.

Thank you.

You can download all jpeg images and origin binary data from here: