MNISTs: All MNIST-like datasets in one package

MNISTs provides an easy way to use MNIST and other MNIST-like datasets (FashionMNIST, KMNIST, EMNIST) in your numpy code.

MNISTs replicates the functionality of torchvision.datasets.mnist without the need to download dozens of dependencies. MNISTs has only one dependency - numpy.

Usage

Each dataset stores train/test images as numpy arrays of shape (n_samples, img_height, img_width) and train/test labels as numpy arrays of shape (n_samples,).

MNIST example:

>>> from mnists import MNIST
>>> mnist = MNIST()
>>> type(mnist.train_images())
<class 'numpy.ndarray'>
>>> mnist.train_images().dtype
dtype('uint8')
>>> mnist.train_images().min()
0
>>> mnist.train_images().max()
255
>>> mnist.train_images().shape
(60000, 28, 28)
>>> mnist.train_labels().shape
(60000,)
>>> mnist.test_images().shape
(10000, 28, 28)
>>> mnist.test_labels().shape
(10000,)
>>> mnist.classes[:3]
['0 - zero', '1 - one', '2 - two']

FashionMNIST example:

from mnists import FashionMNIST
import matplotlib.pyplot as plt

fmnist = FashionMNIST()
plt.imshow(fmnist.train_images()[0], cmap='gray')
plt.title(fmnist.classes[fmnist.train_labels()[0]])
plt.axis('off')
plt.show()

EMNIST example

from mnists import EMNIST
import matplotlib.pyplot as plt

emnist = EMNIST()
letters = emnist.Letters()
plt.imshow(
    letters.train_images()[:256]
        .reshape(16, 16, 28, 28)
        .swapaxes(1, 2)
        .reshape(16 * 28, -1),
    cmap='gray')
plt.axis('off')
plt.show()

Installation

Install mnists from PyPi:

pip install mnists

or from source:

pip install -U git+https://github.com/pczarnik/mnists

The only requirements for MNISTs are numpy>=1.22 and python>=3.9.

If you want to have progress bars while downloading datasets, install with

pip install mnists[tqdm]

Acknowledgments

The main inspirations for MNISTs were mnist and torchvision.datasets.mnist.