i-WiViG: Interpretable Window Vision GNN

This repository contains the implementation of the i-WiViG model from the i-WiViG: Interpretable Window Vision GNN paper.

---

🛠️ Requirements

The project relies on the following key libraries:

• PyTorch
• PyTorch Geometric
• PyTorch Lightning
• wandb (Weights & Biases)

---

🚀 Usage

Training Models

To train the i-WiViG model, as well as the other benchmark models, use the gnn_model_train.py script.

The specific models and datasets are selected by providing input parameters to the primary function:

python gnn_model_train.py 

Generating Explanations

For generating qualitative explanations (e.g., visual subgraphs) and performing the quantitative explanation evaluation, call the edge_attribution.py script:

python edge_attribution.py