An MNIST model deployed online as a full stack application.
Potential Roadmap:
- Deploy as service using Fly.io or github pages
- Further train my model with more epochs and datasets
Front End: Javascript/HTML/CSS
Backend: Python: Flask, Pytorch
Libraries/Dependencies: Ajax, Axios, Matplotlib, PIL
The model is small and simple.
In brief the model used:
-A Convolutional Neural Network (CNN)
-3 epochs with a learning rate of 0.01
-Train batch size of 64, Test batch size of 1000
On the MNIST dataset, the model performed quite well, with a 97% accuracy:
However, the model struggled on the data sent through the web app. Potential issues could be: lack of training, issues with converting HTML5 Canvas images to testing images, or more.
Clone the repo
git clone https://github.com/d0ngeun/Online-MNIST.git
cd Online-MNIST
Train the model
cd ML_Model
python3 model.py
Launch the Flask app
cd..
python -m flask --app server run