XAI Semester Project --- INFO381

This repository contains code and notebooks developed for a semester project on Explainable AI (XAI).

To look at examples of our results, run GUI.py

Project Structure and Execution

The code is designed to be run from the top-level folder INFO381-GITHUB when executing locally.

On Google Colab, this is not an issue as paths are adjusted dynamically.

All notebooks work both locally and in Colab, except for grad-cam.ipynb, which has to be ran locally and not in Colab due to environment compatibility issues.

If you want to run code on Google Colab, you can:

1. upload the code directory to your personal Google Drive
2. identify the path to the code directory (should start with "/content/drive/...")

• Change the argunemt of the cd path to point to where you uploaded the project in your Google Drive. It should look something like /content/drive/MyDrive/INFO381-GitHub

3. double-click the notebook you want to run and replace the string in the Drive connection boilerplate code cell with the path you identified in step 2
4. run the code, and accept the Google Drive pop-u

NOTE: When running on Colab there is no need to unzip the dataset fake_vs_real.zip.

If you want to run code locally:

1. Open top-level folder INFO381-GITHUB in your IDE.
2. We suggest to unzip fake_vs_real.zip manually. The python file utils.py will do it for you, but it takes some time.
3. Run as normal

IMPORTANT TO NOTE:

To use CLIP locally, you need to have an installation of Git or install CLIP manually from https://github.com/openai/CLIP

Folder structure

INFO381-GITHUB/ 
├── gui_images/ # Images used in the GUI 
├── helper_functions/ # Custom helper modules 
│ ├── model_definitions.py # CLIP model architecture definition
│ └── utils.py # Utility functions 
│
├── models/ # Trained models and checkpoints 
│ ├── clip_classifier_10epochs.pth # Trained CLIP model
│ ├── resnet18_cnn.pth # Trained CNN model
│ ├── performance_evaluation.ipynb # Notebook for model evaluation
│ ├── train_CLIP.ipynb # Notebook to train CLIP model
│ └── train_CNN_resnet18.ipynb # Notebook to train CNN model 
│
├── XAI methods/ # Explainable AI method notebooks 
│ ├── grad-cam.ipynb # XAI - Grad-CAM notebook
│ ├── LIME_xai.ipynb # XAI - LIME notebook
│ └── RISE_xai.ipynb # XAI - RISE notebook
│
├── GUI.py # GUI interface 
├── fake_vs_real.zip # Sample dataset (zipped) 
└── README.md