PrinGS: Privacy protection for Gaussian Splatting

PrinGS is a privacy-preserving framework for Gaussian Splatting, with anonymization taking less than 1 second after training. The method locates and anonymizes sensitive regions in the 3D representation based on user reference views, so that renders from new viewpoints contain no sensitive information, while preserving scene geometry and visual fidelity. While retaining a face detection rate FDR of 39.47%, the face similarity FSM is only 9.79%, and the OCR character error rate for sensitive text reaches 1.00, making it completely unreadable.