Load MNIST database in Matlab

mnist dataset of handwritten digits loaded in Matlab

This page explais how to load the MNIST database of handwritten digits with Matlab. This page is based on the converter developed by Markus Mayer, available on Github. If you reached this page, you don't have to download and run the converter, I did it for you. The Markus Mayer's converter output a .mat file containing the data, you can downlod the file below:

Load data

Once the mnist.mat file is downloaded, run the following command to load the dataset:

load ('mnist.mat')

Once the dataset is loaded, type the who command to list the variables:

>> who

Your variables are:

test      training  

The variable training contains the training set, and the variable test contains the test dataset.

mnist dataset of handwritten numbers loaded in Matlab

Each variable is a structure composed of the following elements:

Name Description
count Number of image in the dataset
width Width of each image
height Height of an image
images Data of the images (see below)
labels Label of each image, number written in the image

The images variable is an array of width x height x count the grayscale value of each pixel encoded between 0 and 1.

Display a digits

To check if the dataset is properly loaded, you can display a digit with the following command:

image (training.images(:,:,18)*255);

Since the image is stored with values between 0 and 1, you have to scale the value between 0 and 255. The Matlab function rescale is dedicated to this purpose.

image (rescale(training.images(:,:,18),0,255));

Here is the result:

Display an image of the MNIST dataset with Matlab

To get the save scale on each axis, use the following command:

axis square equal

Display and scale an image of the MNIST dataset with Matlab

To get a gray scale image, use the following:


You can also display the color bar with this command:


Display a grayscale digit of the MNIST dataset with Matlab

Get digit label

Since the label are stored in the dataset, it becomes easy to get the label (or digit) associated to an image:

>> training.labels(18)

ans =


For example, it becomes easy to display the label in the figure title:

title (sprintf ("Digit: %d", training.labels(18)))

Label of the MNIST digit in the Matlab figure title


Last update : 03/02/2022