This is the implementation of the attack proposed by the paper Teacher Model Fingerprinting Attacks Against Transfer Learning
Clone this repo:
git clone https://github.com/yfchen1994/Teacher-Fingerprinting.git
cd Teacher_Fingerprinting
Then install dependencies by:
pip install -r requirements.txt
- MNIST (part of TorchVision)
- CIFAR-10 (part of TorchVision)
- CIFAR-100 (part of TorchVision)
- STL-10 (part of TorchVision)
- VOC-Segmentation (part of TorchVision)
- CelebA. Please download
img_align_celeba
andimg_align_celeba.csv
from this link. Then, create a subfolderceleba
underdataset
, and move the filelist_attr_celeba.csv
and the folderimg_align_celeba
intoceleba
. - Dogs-vs-Cats. Please download the original dataset, unzip
train.zip
, and move the foldertrain
into the./datasets/dogs_vs_cats/
folder. Then run the Python3 scriptprocessdata.py
in./datasets/dogs_vs_cats/
.
The structure of the datasets
folder should be
datasets
+-- dogs_vs_cats
| +-- train
| +-- data_info.csv
+-- celeba
| +-- img_align_celeba
| +-- list_attr_celeba.csv
Please see the example example.sh
.
@inproceedings{CSWZ22,
author = {Yufei Chen and Chao Shen and Cong Wang and Yang Zhang},
title = {{Teacher Model Fingerprinting Attacks Against Transfer Learning}},
booktitle = {{USENIX Security Symposium (USENIX Security)}},
pages = {3593-3610},
publisher = {USENIX},
year = {2022}
}