EMNIST

Overview

The EMNIST (Extended MNIST) dataset is a set of handwritten character digits and letters derived from the NIST Special Database 19. It is designed to follow the exact same file structure and processing as the original MNIST dataset, serving as a drop-in replacement that is significantly harder due to the inclusion of letters.

The dataset includes 6 different splits (configurations). The number of images varies by configuration:

Config	Classes	Train Images	Test Images	Total
digits	10	240,000	40,000	280,000
letters	26	124,800	20,800	145,600
mnist	10	60,000	10,000	70,000
balanced	47	112,800	18,800	131,600
byclass	62	697,932	116,323	814,255
bymerge	47	697,932	116,323	814,255

All images are 28×28 grayscale.

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 handwritten character image
label	int	Class label (Range depends on the selected config)

Usage Example

Basic Usage

You must specify a config_name when loading EMNIST to choose the split (e.g., "digits", "letters", "balanced").

from stable_datasets.images.e_mnist import EMNIST

# Load the 'digits' variant (0-9)
ds_train = EMNIST(config_name="digits", split="train")
ds_test = EMNIST(config_name="digits", split="test")

# Load the 'letters' variant (A-Z)
ds_letters = EMNIST(config_name="letters", split="train")

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

# Optional: make it PyTorch-friendly
ds_train_torch = ds_train.with_format("torch")

References

Official website: https://www.nist.gov/itl/iad/image-group/emnist-dataset

Citation

@misc{cohen2017emnistextensionmnisthandwritten,
  title={EMNIST: an extension of MNIST to handwritten letters},
  author={Gregory Cohen and Saeed Afshar and Jonathan Tapson and André van Schaik},
  year={2017},
  eprint={1702.05373},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/1702.05373},
}