😎This is an official repository of VLAttack: Multimodal Adversarial Attacks on Vision-Language Tasks via Pre-trained Models by Ziyi Yin, Muchao Ye, Tianrong Zhang, Tianyu Du, Jinguo Zhu, Han Liu, Jinghui Chen, Ting Wang and Fenglong Ma.
- We now release the attacking codes of the BLIP and CLIP models [12/10]. More codes for attacking different models are coming soon!
To recreate the environment, run the following command:
$ conda env create -f environment.yaml
$ conda activate VLAttack
$ pip install torch==1.10.0+cu111 torchvision==0.11.0+cu111 torchaudio==0.10.0 -f https://download.pytorch.org/whl/torch_stable.htmlWe independently constructed codes for attacking each model. For example, if you want to test VLAttack on the BLIP model, you need first enter the corresponding directory:
cd BLIP_attack
and then run the commands according to the README.md in the directory.
@InProceedings{VLAttack,
author = {Ziyi Yin and Muchao Ye and Tianrong Zhang and Tianyu Du and Jinguo Zhu and Han Liu and Jinghui Chen and Ting Wang and Fenglong Ma},
title = {VLAttack: Multimodal Adversarial Attacks on Vision-Language Tasks via Pre-trained Models},
booktitle = {NeurIPS},
year = {2023}
}
VLAttack is released under BSD 3-Clause License. Please see LICENSE file for more information.