Access Control of Semantic Segmentation Models Using Encrypted Feature Maps

In this paper, we propose an access control method with a secret key for semantic segmentation models the first time so that unauthorized users without cannot benefit from performance of trained models. The enables us not only to provide high authorized but also degrade users. We point out that, application segmentation, conventional methods which use encrypted images classification tasks are directly applicable due degradation. Accordingly, in selected feature maps training and testing models, instead input images. experiment, protected allowed obtain almost same as non-protected robustness against key.