This repository contains code to reproduce results from the paper:

Triangle Attack: A Query-efficient Decision-based Adversarial Attack (ECCV 2022)

Xiaosen Wang, Zeliang Zhang, Kangheng Tong, Dihong Gong, Kun He, Zhifeng Li, Wei Liu

• python >= 3.6.5
• pytorch == 1.7.x
• numpy >= 1.15.4
• imageio >= 2.6.1
• torch_dct >= 0.1.5

Firstly, you should prepare your own benign images and victim models for attack. The pathes for the input images and model are set by --dataset_path and --modelpath, respectively. You could also download our sampled 200 images used in the experiments and adopt the the pretrained models in pytorch.

You could run TA as follows:

CUDA_VISIBLE_DEVICES=gpuid python TA.py --dataset_path images --csv label.csv

The generated adversarial examples would be stored in directory ./output_folder. We report the attack success rates under the thresholds of 0.1, 0.05 and 0.01 respectively.

If you find the idea or code useful for your research, please consider citing our paper:

@inproceedings{wang2022Triangle,
  author={Xiaosen Wang and Zeliang Zhang and Kangheng Tong and Dihong Gong and Kun He and Zhifeng Li and Wei Liu},
  booktitle = {European Conference on Computer Vision},
  title = {Triangle Attack: A Query-efficient Decision-based Adversarial Attack},
  year = {2022},
}

Questions and suggestions can be sent to [email protected].