RISE-CAM: Explainable AI for Image Classification

This repository contains source code necessary to reproduce the main results in a Final Year Project finished at the City University of Hong Kong.

• Student name: WU, Chou-Ling
• Student number: 56512971
• Project code: 23CS029
• Project title: Explainable AI (XAI) for Image Classification, Final Report

Method overview

This repository introduces two novel algorithms--RISE-CAM and RISExCAM. These algorithms are merge from two classic XAI algorithms, RISE and Grad-CAM. Below is the flow chart of how RISE-CAM works.

Repository contents

1. Introduction_RISECAM.ipynb is a brief tour to go through the mechanisms of RISE and Grad-CAM, and how RISE-CAM was developed based on them. This notebook demonstrates the outputs of each algorithms.
2. Evaluation_RISECAM.ipynb provides a thorough comparison of RISE, Grad-CAM, RISE-CAM, and RISExCAM. The there are different evaluation metrics used here. This file is too large to be previewed on GitHub.
3. Evaluation_RISECAM_lite.ipynb is a notebook without the output saliency maps and with a smaller file size. Thus, it can be previewed.
4. Most of the functions used in this project can be found in algorithms.py.

Before you start...

Install the required packages.

pip install -r requirements.txt

The images in the "examples" folder serve the prupose of simple demonstration. To perform a complete evaluation, please download the dataset from:

https://www.kaggle.com/datasets/iamsouravbanerjee/animal-image-dataset-90-different-animals

Remember to unzip the folder under this directory before running Evaluation_RISECAM.ipynb.

Examples