Digit Recognizer — MNIST

Classify handwritten digits (0-9) from 28x28 grayscale images.

Computer VisionCNN

Model Result

A convolutional neural network trained in Keras. The exact score is produced in the training notebook run; this showcase focuses on the approach and the live demo.

Key Insight

A compact convolutional network with augmentation reaches ~0.996 accuracy; test-time augmentation trims the last few misclassifications.

Try the live app View on Kaggle