number-classifier web application

Technologies used

• PyTorch
• MongoDB (Mongoengine)
• Django

Aim

The aim of this project was to get familiar with neural networks as well as practice using a backend framework (Django) and an ORM (mongoengine)

Challenges

The biggest challenge encountered was implementing the neural network. At first a classic feed forward neural network was used, however the prediction accuracy was quite low (about 67%). To fix this, a convolutional neural network was implemeted and the prediction accuracy was much better (about 87%).

The Project

This is a fairly simple project, you can create a new user account, login as an existing user, upload and submit images (hand written numbers). Once the user has submitted the image, a prediction of what the number submitted is printed out on the screen.

How to run the project

1. Ensure you have PyTorch, Torchvision, mongoengine and Django packages installed
2. In the same directory as manage.py execute the following command: python manage.py runserver