phoenixwilliams/Black-Box-Sparse-Adversarial-Attack-via-Multi-Objective-Optimisation

Code for our CVPR 2023 paper: Black-Box Sparse Adversarial Attack via Multi-Objective Optimisation

Python

Black-Box Sparse Adversarial Attack via Multi-Objective Optimisation

This repository contains the code for our 2023 CVPR paper "Black-Box Sparse Adversarial Attack via Multi-Objective Optimisation"

Installation


1. Download repository 

2. pip install -r requirements.txt

Attacking models


- The method relies on a taking a callable function that returns
the loss off of an adversarial image i.e. f(x_adv}). The method assumes the task is minimization problem. 

- View main.py for an example of how to run the method. We provide the suggested parameters there.

Citation

@inproceedings{williams2023black,
  title={Black-Box Sparse Adversarial Attack via Multi-Objective Optimisation},
  author={Williams, Phoenix Neale and Li, Ke},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={12291--12301},
  year={2023}
}

Paper


You can access the paper [here].