DigitRecognizer

A simple React app to recognise handwritten numeric digits using a pre-trained CNN.

This web app has been deployed using Netlify here. Make sure you check it out!

About the Project

Recognizing Handwritten digits using a CNN Model trained on the MNIST Dataset is a milestone for many Machine Learning beginners (including myself XD). There are many models available on the internet with more than 99.7% accuracy, and my model is no match to those! I wanted to create a simple project which does the same task of recognizing digits, but now it is an interactive app where users can draw their digits and my model recognizes them. This project was the inspiration behind making this web app.

Convolutional Neural Network

I have used a 3-layer deep Convolutional Neural Network and trained it on the MNIST Dataset with an accuracy of 99.3%. This model is saved in tfjs format using the Python tfjs API.

React frontend

I have designed the web application using React and React Bootstrap libraries. The frontend uses the saved CNN model to predict the user input and displays it.

Getting Started

To run this app,

• Clone this repository

  git clone https://github.com/GauthamBellamkonda/DigitRecognizer.git

• Install node packages/dependencies

  npm install

• Run this app

  npm run start

TO DO

• Write a decent README
• Make a chart of predictions
• Cleanup/Refactor code
• WIP: Improve the model
  • Use TensorBoard to monitor the accuracy/results of the model
• Add colors to frontend, make it more interactive/beautiful
• Explain the complete working/design of the app either in README or the page
• Minor changes in CNN
• Recognize multiple digits in one shot
• Recognize text (all alphanumeric characters)