MoreBayesian-attack

Code for our ICLR 2023 paper Making Substitute Models More Bayesian Can Enhance Transferability of Adversarial Examples.

Requirements

• Python 3.8.8
• PyTorch 1.12.0
• Torchvision 0.13.0

Datasets

Select images from ImageNet validation set, and write .csv file as following:

class_index, class, image_name
0,n01440764,ILSVRC2012_val_00002138.JPEG
2,n01484850,ILSVRC2012_val_00004329.JPEG
...

Finetune, Attack, and Evaluate

Finetune

Perform our finetune with SWAG:

python3 finetune.py --data_path ${IMAGENET_DIR} --save-dir ${MODEL_SAVE_DIR}

You can download our finetuned ResNet-50 at Google Drive.

Attack

Perform attack:

python3 attack.py --source-model-dir ${SOURCE_MODEL_DIR} --data-dir ${IMAGENET_VAL_DIR} --data-info-dir ${DATASET_CSV_FILE} --save-dir ${ADV_IMG_SAVE_DIR}

Evaluate

Evaluate the success rate of adversarial examples:

python3 test.py --dir ${ADV_IMG_SAVE_DIR} --model_dir ${VICTIM_MODEL_WEIGHTS_DIR}

Acknowledgements

The following resources are very helpful for our work:

• Pretrained models for ImageNet

Citation

Please cite our work in your publications if it helps your research:

@article{li2023making,
  title={Making Substitute Models More Bayesian Can Enhance Transferability of Adversarial Examples},
  author={Li, Qizhang and Guo, Yiwen and Zuo, Wangmeng and Chen, Hao},
  booktitle={ICLR},
  year={2023}
}