MNISET

Small MNIST like dataset with +- 4000 28x28 RGB SET card game images, with different attributes for a total of 3 * 3 * 3 * 3 = 81 classes:

3 different quantities: 1, 2 and 3
3 different fills: open, striped and solid
3 different colors: red, green and purple
3 different shapes: stadium, diamond and squiggle

Dataset gathered as a side-effect of developing the SET Finder app.

Color MNISET

The full color dataset has 3^4 = 81 classes and 28x28x3 (RGB) color images. See below some example images or see the full train set or full test set.

Usage

There is no need to clone this repository, simply include below snippet of code:

import os
# Not that pretty but this way you just need this code
if not os.path.isfile('mniset.py'):
    from urllib import request
    request.urlretrieve('https://github.com/wouterkool/MNISET/raw/master/mniset.py', 'mniset.py')

from mniset import load_mniset, extract_dataset, extract_grayscale_dataset
mniset = load_mniset()
x_train, y_train, labels, label_imgs = extract_dataset(mniset, split='train')
x_test, y_test, *_ = extract_dataset(mniset, split='test')

Grayscale MNISET

The default grayscale dataset has 3^3 = 27 classes and 28x28 grayscale images. Below you can see an example of pictures for each class (with an 'image label column' on the left):

The classes come in different frequencies:

Usage

There is no need to clone this repository, simply include below snippet of code:

import os
# Not that pretty but this way you just need this code
if not os.path.isfile('mniset.py'):
    from urllib import request
    request.urlretrieve('https://github.com/wouterkool/MNISET/raw/master/mniset.py', 'mniset.py')

from mniset import load_mniset, extract_grayscale_dataset
mniset = load_mniset()
x_train, y_train, labels, label_imgs = extract_grayscale_dataset(mniset, split='train')
x_test, y_test, *_ = extract_grayscale_dataset(mniset, split='test')

Other possibilities

The dataset can also be used for 3 or 9 class classification problems, both with the color or grayscale dataset. For more details, as well as the plotting functions used here, see MNISET.ipynb.

Disclaimer

This dataset is under development and may change without notice. It may contain a few mistakes. Contact me if you have any questions.
SET is a registered trademark by Cannei, LLC. This repository is not in any way affiliated.