Welcome to our repository featuring the official implementation of the CGBA algorithm. This work is accepted for publication in ICCV 2023.
The arXiv version of the paper is available here.
Before executing the code, ensure that the following packages are installed in your environment:
- PyTorch and Trochvision
- Numpy
- Os
- SciPy
Or you can type the following to create an environment:
conda env create -f cgba_env.yml
- To execute the non-targeted attack, run
Non_targeted_attack.py
. - To execute the targeted attack, run
Targeted_attack.py
.
Please consider citing our paper in your publications if it contributes to your research.
@InProceedings{Reza_2023_ICCV,
author = {Reza, Md Farhamdur and Rahmati, Ali and Wu, Tianfu and Dai, Huaiyu},
title = {CGBA: Curvature-aware Geometric Black-box Attack},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month = {October},
year = {2023},
pages = {124-133}
}