MNIST Explorer - About

Draw a digit and see a simple MNIST classifier network process it in real-time.

The model features a 28x28 input array, and two dense layers (128 and 10)

This model was trained on the classic MNIST dataset. The model mis-predicts sometimes on our canvas, since we have done no further transfer learning, however this is intended.

Motivation

This project was completed in two weeks during June 2021 as an exploration into team dynamics.

Technologies Used

SvelteJS, p5.js, TensorFlow/Keras, tf.js, deployed on Netlify

Source Code

Credits

MNIST Explorer

Prediction: Unsure
Confidence: 36.9%

Model Output:

0=0.04 1=0.05 2=0.07 3=0.07 4=0.10 5=0.37 6=0.06 7=0.10 8=0.07 9=0.07

Draw a digit: