RPA

About

Corresponding code to the paper "Enhancing the transferability of adversarial examples with ramdom patch" (IJCAI 2022).

Quick Start

Prepare pretrained models

• Normlly trained models
• Adversarial trained models
• Visual transformers models

Running attacks

• RPA

python attack.py --model_name vgg_16 --attack_method RPA --layer_name vgg_16/conv3/conv3_3/Relu --ens 60 --probb 0.7

Evaluate the success rate

• Normlly trained models nad Adversarial trained models

python verify_tf_model.py

• Visual transformers models

python verify_vits.py

Acknowledgments

Code refers to FIA

Citing this work

@inproceedings{zhang2022enhancing,
  title={Enhancing the transferability of adversarial examples with ramdom patch},
  author={Zhang, Yaoyuan and Tan, Yu-an and Chen, Tian and Liu, Xinrui and Zhang, Quanxin and Li, Yuanzhang},
  booktitle={IJCAI},
  pages={1672-1678},
  year={2022}
  }