Visualization-Capsule-Neural-Networks

About the Project

This project is developed as part of my Bachelor's degree at Technical University in Kosice, under the guidance of M.Eng. Dominik Vranay, PhD. and my supervisor prof. Ing. Peter Sinčák, CSc.

Project Status

⚠️ This project is currently in development. The functionality and features provided are stable, but the project is being actively improved.

Overview

This repository contains a Python implementation of a Capsule Neural Network (CapsNet) that is trained and evaluated on the MNIST dataset. The project focuses on visualizing the coupling coefficients of the network, leveraging a GraphVisualizer class that represents the connections between capsules.

Features

• Training and evaluation of a Capsule Neural Network using PyTorch.
• Custom class MySampler for balancing classes in each batch during training.
• Visualization of coupling coefficients with an interactive graph.
• A web application built with Dash for an interactive user experience.

Requirements

• Python
• PyTorch
• NetworkX
• Dash
• Plotly
• Matplotlib

Installation

Clone the repository and install the required packages:

git clone https://github.com/LordWhiskas/Visualization-Capsule-Neural-Networks.git

cd Visualization-Capsule-Neural-Networks

pip install -r requirements.txt

Using

Run the file modelLoad.py

Go to the url that you will see in console

Click button update graph

You can change image using right or left buttons

After changing image click on update graph to see new graph

Changing model

❗Use custom sampler MySampler for balancing classes in your dataset. It will improve your model's accuracy.

Acknowledgments

I would like to express my deepest appreciation to Dominic Vranay, who is providing expert guidance and support throughout the development of this project.