Dataset · ylecun
MNIST 28x28 handwritten digits
ylecun/mnist
70k black-and-white 28x28 handwritten digit images for training and validating classifiers.
What's inside
This consists of 70,000 grayscale handwritten digit images of 28x28 pixels across 10 classes, with 60,000 images in the training set and 10,000 in the validation set (7,000 per class). The data is extracted from two NIST databases: half of the digits were written by Census Bureau employees and half by high school students. The metadata language is English.
What you can do with it
- Train a CNN or MLP for handwritten digit classification
- Test an OCR pipeline on digit recognition
- Use as an image-classification benchmark
- Tasks
- Image classification
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 70.3K
- Likes
- 302
Models trained on it
- asnelt/mnistvit-transformersImage classification 538
- building-an-inference-engine-from-scratch/mnist-tfliteImage classification 161
- prithivMLmods/Mnist-Digits-SigLIP2Image classification 102
- tanganke/clip-vit-base-patch32_TA8_mtl 101
- eoinedge/TinyDigitNetImage classification 90
- kenil-patel-183/mnist-cnn-digit-classifierImage classification 65
- chrisjcc/mnist-cnnImage classification 65
- chrisjcc/mnist-vitImage classification 56
How to load it
from datasets import load_dataset
ds = load_dataset("ylecun/mnist")