README.md

metadata

license: mit

AlphaNum Dataset

Dataset Summary

The AlphaNum dataset, created by Louis Rädisch, is a collection of handwritten grayscale characters and digits of the size 28x28 intended for the use in Optical Character Recognition (OCR) tasks. This dataset is a compilation from various sources and includes labels from 0-35, where 0-25 represent the letters A-Z and 26-35 represent the digits 0-9. The images from the MNIST dataset included have been color inverted. To bring together the different datasets, vision transformer Models have been fine tuned to further increase the accuracy of the data. For instance the "A-Z handwritten alphabets" had no seperation between big and small letters.

Sources:

1. https://github.com/sueiras/handwritting_characters_database
2. mnist
3. https://www.kaggle.com/datasets/sachinpatel21/az-handwritten-alphabets-in-csv-format
4. https://www.kaggle.com/datasets/thomasqazwsxedc/alphabet-characters-fonts-dataset

Dataset Structure

Data Instances

A data instance in this dataset is an image of a handwritten character or digit along with its corresponding label.

Data Fields

1. 'image': The image of the handwritten character or digit.
2. 'label': The label for the character or digit in the image. (0-25 for A-Z, 26-35 for 0-9)

Data Splits

The dataset is divided into training and test sets.

Dataset Use

The AlphaNum dataset is suitable for tasks related to text recognition, document processing, and machine learning, particularly in the development and improvement of OCR models.