This repository contains the PyTorch implementation of the following paper to appear at ICML 2024: "Intersecting-Boundary-Sensitive Fingerprinting for Tampering Detection of DNN Models".
First, install all dependencies via pip.
$ pip install -r requirements.txt
Second, prepare dir to save fingerprints
$ mkdir outputs
$ cd outputs
$ mkdir fingerprints
$ cd ../..
Below is a demo to generate 1000 fingerprint samples on GPU device
$ python mian.py --gpu '1' --num 1000 --dataset 'cifar10'
This saves the generated fingerprints file at outputs/fingerprints/cifar10
.
Our IBSF currently implements custom data loaders for the following datasets.
- CIFAR-10
- ImageNet (needs manual download)
- GTSRB (needs manual download)
@inproceedings{
xiaofan2024intersectingboundarysensitive,
title={Intersecting-Boundary-Sensitive Fingerprinting for Tampering Detection of {DNN} Models},
author={Bai Xiaofan and Chaoxiang He and Xiaojing Ma and Bin Benjamin Zhu and Hai Jin},
booktitle={Forty-first International Conference on Machine Learning},
year={2024},
url={https://openreview.net/forum?id=kZArjKc64o}
}