Figure 1:GE-AdvGAN Schematic Diagram |
This repository contains the official implementation of the GE-AdvGAN, as presented in the paper GE-AdvGAN: Gradient Editing-based Adversarial Generative Model. The GE-AdvGAN framework is designed to enhance the transferability of adversarial examples through a novel approach that involves editing gradients within an adversarial generative model context.
Before running the experiments, ensure that your environment meets the following prerequisites:
- Python version: 3.8
- PyTorch version: 1.8
- Pretrainedmodels: 0.7
- NumPy version: 1.19
- Pandas version: 1.2
Please install the required libraries using pip
or a similar package manager to meet the above specifications.
The pretrained models necessary for running the experiments can be downloaded from the following link:
Ensure that you place the downloaded models in the appropriate directory within your project structure.
To facilitate the execution of experiments, we provide shell scripts for both the baseline and the GE-AdvGAN experiments.
To run the baseline experiment, execute the following command:
sh run_baseline.sh
To conduct the GE-AdvGAN experiment, use the following command:
sh run_GE.sh
You are encouraged to modify the parameters within these shell scripts to tailor the experiments to your specific requirements.
@article{zhu2024ge,
title={GE-AdvGAN: Improving the transferability of adversarial samples by gradient editing-based adversarial generative model},
author={Zhu, Zhiyu and Chen, Huaming and Wang, Xinyi and Zhang, Jiayu and Jin, Zhibo and Choo, Kim-Kwang Raymond},
journal={arXiv preprint arXiv:2401.06031},
year={2024}
}
Code refer to: advGAN_pytorch