Explainability and Robustness in Metric Space

This repository implements explainability and robustness methods such as Saliency Maps, Integrated Gradients, FGSM and PGD Attacks applied to point cloud classification models.

Installation

• Set up and activate conda environment.

conda env create -f environment.yml
conda activate erms

• Install pre-commit hooks.

pre-commit install

• Download pretrained checkpoint.

mkdir out
gdown --no-cookies --id 1-2UrF5V_gpjGNbWfbk_kEp742Bnjlnc- -O out/pointmlp.pth

Quick Start

• Run all experiments and plot results.

python run.py