Dispersion Reduction Attack

Code for CVPR2020 paper Enhancing Cross-task Black-Box Transferability of Adversarial Examples with Dispersion Reduction.

Docker Quick Start

Build the docker image and all dependencies will be installed automatically.

docker pull erbloo/baidu_aisec:yantao_v1.0.0

Get access to the container and clone the repository into it using command:

docker exec -ti |container_name| bash
git clone https://github.com/erbloo/dr_cvpr20.git

Experimental Data

Experimental data is included in dr_images_cvpr20.

Please find the attack example in example.ipynb

Citing this work

If you find this work is useful in your research, please consider citing:

    @inproceedings{lu2020enhancing,
      title={Enhancing cross-task black-box transferability of adversarial examples with dispersion reduction},
      author={Lu, Yantao and Jia, Yunhan and Wang, Jianyu and Li, Bai and Chai, Weiheng and Carin, Lawrence and Velipasalar, Senem},
      booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
      pages={940--949},
      year={2020}
    }