Kuzushiji-MNIST

Overview

The Kuzushiji-MNIST dataset is a widely-used image classification benchmark for single-label classification consisting of 70,000 28×28 grayscale images of cursive Japanese characters across 10 balanced classes (o, ki, su, tsu, na, ha, ma, ya, re, wo), with 7,000 images per class.

• Train: 60,000 images (6,000 per class)

• Test: 10,000 images (1,000 per class)

Data Structure

When accessing an example using ds[i], you will receive a dictionary with the following keys:

Key	Type	Description
image	PIL.Image.Image	28×28 grayscale image
label	int	Class label (0-9)

Usage Example

Basic Usage

from stable_datasets.images.k_mnist import KMNIST

# First run will download + prepare cache, then return the split as a HF Dataset
ds = KMNIST(split="train")

# If you omit the split (split=None), you get a DatasetDict with all available splits
ds_all = KMNIST(split=None)

sample = ds[0]
print(sample.keys())  # {"image", "label"}

# Optional: make it PyTorch-friendly
ds_torch = ds.with_format("torch")

Related Datasets

• MNIST: Original handwritten digit recognition dataset

• Fashion-MNIST: Drop-in replacement for MNIST with fashion items

• EMNIST: Extended MNIST with letters and digits

• Kuzushiji-49: Extended version with 49 classes of Kuzushiji characters

References

• Official website: http://codh.rois.ac.jp/kmnist/

• License: CC BY-SA 4.0

Citation

@online{clanuwat2018deep,
author       = {Tarin Clanuwat and Mikel Bober-Irizar and Asanobu Kitamoto and Alex Lamb and Kazuaki Yamamoto and David Ha},
title        = {Deep Learning for Classical Japanese Literature},
date         = {2018-12-03},
year         = {2018},
eprintclass  = {cs.CV},
eprinttype   = {arXiv},
eprint       = {cs.CV/1812.01718}
}