Language Guided Adversarial Purification

The attacks can be evaluated using

• preprocessor_blind.py
• robust_acc_bpda_eot.py

Update the dataset and the classifier model accordingly.

Notebooks for training and evaluation are provided in notebooks folder.

Pretrained model can be downloaded here

Comments

This code is adapted from

• https://github.com/CompVis/stable-diffusion
• https://github.com/salesforce/BLIP
• https://github.com/jmyoon1/adp

BibTeX

@article{singh2023language,
  title={Language Guided Adversarial Purification},
  author={Singh, Himanshu and Subramanyam, AV},
  journal={arXiv preprint arXiv:2309.10348},
  year={2023}
}