Awesome
AdvDiff ECCV 2024
The code repository for our paper AdvDiff: Generating Unrestricted Adversarial Examples using Diffusion Models arXiv
Installation
This repository is based on the official code from Latent Diffusion Models.
-
Set up environments for the codes. Details please refer to the original Github code.
conda env create -f environment.yaml conda activate ldm_adv
-
Download ImageNet checkpoint (https://ommer-lab.com/files/latent-diffusion/nitro/cin/model.ckpt) and save to "models/ldm/cin256-v2/model.ckpt"
Usage
Simply run advdiff.py to perform attacks against the official PyTorch ResNet50 model. You can modify the attack parameter at the advdiff.py.
python advdiff.py
Reference
Please cite our paper if you found any helpful information:
@article{dai2023advdiff,
title={AdvDiff: Generating Unrestricted Adversarial Examples using Diffusion Models},
author={Dai, Xuelong and Liang, Kaisheng and Xiao, Bin},
journal={arXiv preprint arXiv:2307.12499},
year={2023}
}