riolys / disdiff

Disrupting Diffusion: Token-Level Attention Erasure Attack against Diffusion-based Customization(ACM MM2024)

DisDiff (ACM MM2024) 🌟

Official Code of Disrupting Diffusion: Token-Level Attention Erasure Attack against Diffusion-based Customization(ACM MM2024)

Environment

Please follow Anti-DreamBooth to install the environment&base codes&datasets.

For visualizations, use this notebook of Attend and excite.

Codes and Usage

Concat the attacks and scripts folders with the same name folders in Anti-DreamBooth.

Then conduct

bash scripts/disdiff.sh 

Cite

If you find DisDiff is helpful to your work, please cite our work

@article{liu2024disrupting,
  title={Disrupting Diffusion: Token-Level Attention Erasure Attack against Diffusion-based Customization},
  author={Liu, Yisu and An, Jinyang and Zhang, Wanqian and Wu, Dayan and Gu, Jingzi and Lin, Zheng and Wang, Weiping},
  journal={arXiv preprint arXiv:2405.20584},
  year={2024}
}

disdiff's Issues

Missing detailed environment configuration and some necessary running modules

Dear authors, I meet the following problems:

1. train_dreambooth.py used in disdiff.sh is lacked.
2. Line 32 of disdiff.py, "from show_attention import Attmap" causes an error "ModuleNotFoundError: No module named 'show_attention", I cannot find this module from your codes.
   I also suggest you provide detailed environment configuration in this repository, such as a requirements.txt.