This code is similar to save cifar10 one but in this code we use the tensorflow methods that extarct images and extract labels then we reshape, transpose and save the images in intended folders to be ready for keras users.

The output of this code is perfect for keras users that uses the flow from directory method

Dataset preparation is the hardest step in machine learning. it takes 90% of effort in the process on machine learning.

After a long search for code that converts the binary files of mnist to jpg images and save them in folders, nothing was found.

EMNIST binary files are the same as mnist, so this code is applicable to be applied for any type of mnist version of 10 classes

if you want to apply it for more than 10 classes you need to edit it, it is easy and clear.

http://yann.lecun.com/exdb/mnist/