binarized_mnist

A specific binarization of the MNIST images originally used in (Salakhutdinov & Murray, 2008). This dataset is frequently used to evaluate generative models of images, so labels are not provided.

Split Examples
'test' 10,000
'train' 50,000
'validation' 10,000
  • Feature structure:
FeaturesDict({
    'image': Image(shape=(28, 28, 1), dtype=uint8),
})
  • Feature documentation:
Feature Class Shape Dtype Description
FeaturesDict
image Image (28, 28, 1) uint8

Visualization

  • Citation:
@inproceedings{salakhutdinov2008quantitative,
title={On the quantitative analysis of deep belief networks},
author={Salakhutdinov, Ruslan and Murray, Iain},
booktitle={Proceedings of the 25th international conference on Machine learning},
pages={872--879},
year={2008},
organization={ACM}
}