digit-recognition-with-convolutional-neural-networks

This project recognises digits using convolutional neural network. It is trained on the MNIST dataset. I am working on making the recognition custom. This will be updated soon

This project is configured for Linux and uses python3

To run this project, open up your bash terminal and write

chmod -R 777 setup.sh
./setup.sh

This will set up the project enviornment for you. This must be run with administrator rights. After you set up the project, run:

sudo python3 run_me.py

OR

*** virtual enviornment will be setup through setup.sh ***

source venv/bin/activate
pip install -r requirements.txt
python run_me.py

The following images should guide you with the rest of the process

Run Project

Training

Training Accurracy

Testing

Generate graphs

Graphs

Accuracy vs Iterations with 0.001 lr, 0.75 dropout and 250000 Iterations

Loss vs Iterations with 0.001 lr, 0.75 dropout and 250000 Iterations

Time taken vs Iterations with 0.001 lr, 0.75 dropout and 250000 Iterations

Author: Satyaki Sanyal

This project is strictly for educational purposes only.