This repository is meant to provide a version of MNIST dataset, which is shadowed by some percentages with randomly created masks.
To recreate the dataset, run
python creation.py
You might need to install some dependencies (provided at requirements.txt).
To use the dataset, read it with the following python code (or any equivalent in another language):
import numpy as np
images = np.fromfile('./data/train/shadow=0.9/images.bin', dtype=np.bool)