MNIST-100

This dataset is a specialized subset of the renowned dataset, designed for focused research and applications in digit recognition. This dataset narrows down the original scope to feature 100 distinct handwritten numbers, offering a unique tool for machine learning practitioners and researchers.

Dataset Specifications:

• Name: MNIST-100
• Total Images:
  • Training Images: 60,000
  • Test Images: 1,000
• Classes: 100 (Featuring numbers from 00 to 99)
• Image Dimensions: 28×56 pixels (Grayscale)

Data Compilation Methodology: The Dataset was constructed by extracting 10 unique digits from the original  database. From each of these digits, 10 representative images were selected to ensure a diversity in handwriting styles. This process resulted in a curated collection of 100 images, each labeled with a two-digit number ranging from 00 to 99.

Purpose and Utility: The Dataset serves as a supplementary dataset to the traditional, tailored for experiments requiring detailed analysis on a compact scale. It is ideal for training and testing machine learning models that perform digit classification, particularly in scenarios where a more manageable dataset can facilitate quicker iterations and more granular insight.