Membership Inference Attacks are Easier on Difficult Problems
Membership Inference Attacks are Easier on Difficult Problems
Avital Shafran, Shmuel Peleg and Yedid Hoshen
International Conference on Computer Vision (ICCV), 2021.
Usage
The current software is tested with Pytorch 1.6.0 and Python 3.6.
Dataset
Download the Cityscapes dataset from the official website (registration required).
After downloading, please put it under the ./pix2pixHD/datasets folder in the same way the example images are provided.
Pre-trained model
Please download the pre-trained pix2pixHD Cityscapes model from the official pix2pixHD Pytorch implementation, and put it under ./pix2pixHD/checkpoints/label2city_1024p/
Run attack
Run attack on pre-trained pix2pixHD model:
python run_MIA_pix2pixHD.py --dataroot./pix2pixHD/datasets/Cityscapes --checkpoints_dir ./pix2pixHD/checkpoints
Citation
If you find this project useful for your research, please cite
@article{shafran2021reconstruction,
title={Reconstruction-Based Membership Inference Attacks are Easier on Difficult Problems},
author={Shafran, Avital and Peleg, Shmuel and Hoshen, Yedid},
journal={arXiv preprint arXiv:2102.07762},
year={2021}
}
Acknowledgments
This code borrows heavily from pix2pixHD.