Model Barrier: A Compact Un-Transferable Isolation Domain for Model Intellectual Property Protection

Code release for "Model Barrier: A Compact Un-Transferable Isolation Domain for Model Intellectual Property Protection" (CVPR 2023)

Paper

Model Barrier: A Compact Un-Transferable Isolation Domain for Model Intellectual Property Protection (CVPR 2023)

We propose a Compact Un-Transferable Isolation Domain (CUTI-domain), which acts as a barrier to block illegal transfers from authorized to unauthorized domains, to protect the intellectual property (IP) of AI models.

Prerequisites

The code is implemented with CUDA 11.4, Python 3.8.5 and Pytorch 1.8.0.

Datasets

MNIST

MNIST dataset can be found here.

USPS

USPS dataset can be found here.

SVHN

SVHN dataset can be found here.

MNIST-M

MNIST-M dataset can be found here.

CIFAR-10

CIFAR-10 dataset can be found here.

STL-10

CIFAR-10 dataset can be found here.

Office-Home

Office-Home dataset can be found here.

VisDA 2017

VisDA 2017 dataset can be found here.

Running the code

Target-Specified CUTI-Domain

python train_ts_dight.py

Ownership Verification

python train_owner_dight.py

Target-free CUTI-Domain

python train_tf_dight.py

Applicability Authorization

python train_author_dight.py

Citation

If you find this code useful for your research, please cite our paper:

@inproceedings{wang2023model,
  title={Model barrier: A compact un-transferable isolation domain for model intellectual property protection},
  author={Wang, Lianyu and Wang, Meng and Zhang, Daoqiang and Fu, Huazhu},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={20475--20484},
  year={2023}
}

Acknowledgements

Some codes are adapted from NTL and SWIN-Transformer. We thank them for their excellent projects.

Contact

If you have any problem about our code, feel free to contact

• lywang12@126.com
• wangmeng9218@126.com

or describe your problem in Issues.