Try the Live Demo

mnist-draw

This repository contains a single page website that enables users to draw digits (0-9) by hand and have the digits classified in real time as they are drawn by a machine learning model trained against the MNIST dataset.

Setup

Python 3.5+ is required for compatibility with all required modules

# Clone this repository
git clone https://github.com/marcacohen/mnist-draw.git

# Change to the repo root directory
cd mnist-draw

# Install required modules
pip install -r requirements.txt

Usage

To launch the website, begin by starting a Python server from the repository folder:

# Start Python server (use Python 3)
python -m http.server --cgi 8000

Then open a browser and navigate to http://localhost:8000/.

An example of the website's interface is shown below. Users are guided to draw a digit (0-9) on the empty canvas and watch the page try to classify your digit as you draw.

Results are displayed as a bar graph where each classification label recieves a score between 0.0 and 1.0 from the machine learning model. Clear the canvas with the 'Clear' button to draw and process other digits.

Interface example:

Machine Learning Model

The model I'm using for this page a convolutional neural network (CNN) built using Keras/TensorFlow on a Google Tensor Processing Unit (TPU). The model is explained in depth and is available for you to build yourself here.

Acknowledgements

Thanks to rhammell for the original Python version and to zackakil for adding support for Tensorflow.js. In this version I added predicting while you draw, deployed the example UI on Cloud Run (bit.ly/mco-mnist-draw), and deployed a Colab notebook (bit.ly/mco-mnist-lab) so you can build your own version of this model, with the UI widget integrated right into the notebook.