An Access Control Method with Secret Key for Semantic Segmentation Models

TL;DR

This paper introduces a secret key-based access control method for semantic segmentation models using vision transformers, enabling secure model usage and maintaining accuracy for authorized users while degrading it for unauthorized ones.

Contribution

It presents a novel encryption-based access control approach tailored for ViT-based segmentation models, addressing limitations of prior CNN-focused methods.

Findings

Authorized users achieve the same accuracy as with plain images.

Unauthorized users experience significantly degraded accuracy.

The method effectively protects models from unauthorized access.

Abstract

A novel method for access control with a secret key is proposed to protect models from unauthorized access in this paper. We focus on semantic segmentation models with the vision transformer (ViT), called segmentation transformer (SETR). Most existing access control methods focus on image classification tasks, or they are limited to CNNs. By using a patch embedding structure that ViT has, trained models and test images can be efficiently encrypted with a secret key, and then semantic segmentation tasks are carried out in the encrypted domain. In an experiment, the method is confirmed to provide the same accuracy as that of using plain images without any encryption to authorized users with a correct key and also to provide an extremely degraded accuracy to unauthorized users.