/MMP-Attack

Official repository for "On the Multi-modal Vulnerability of Diffusion Models"

Primary LanguagePython

MMP-Attack

This repository is the official repository for our paper "On the Multi-modal Vulnerability of Diffusion Models"

bounding box

Setup

Install PyTorch, diffusers==0.17.0, and transformers==4.29.1.

Download the Stable Diffusion model and CLIP model.

Running Commands

You can execute the following command to perform a targeted attack, transforming the category "car" into "bird":

python attack.py --ori_sentence "a photo of car" --target_word bird

Citation

If you benefit from our work in your research, please consider to cite the following paper:

@inproceedings{
yang2024on,
title={On the Multi-modal Vulnerability of Diffusion Models},
author={Dingcheng Yang and Yang Bai and Xiaojun Jia and Yang Liu and Xiaochun Cao and Wenjian Yu},
booktitle={Trustworthy Multi-modal Foundation Models and AI Agents (TiFA)},
year={2024},
url={https://openreview.net/forum?id=FuZjlzR7kT}
}

Please feel free to contact us if you have any questions.