Privatar: Scalable privacy-preserving multi-user VR through secure offloading
Privatar: Scalable Privacy-preserving Multi-user VR via Secure Offloading
AI summary
Privatar offloads multi-user VR avatar reconstruction from the headset to untrusted devices within the local network, supporting 2.37× concurrent users on the Meta Quest Pro at a cost of 6.5% higher reconstruction loss and 9% energy overhead. Its Horizontal Partitioning offloads only low-energy frequency components, and DAMP minimizes noise according to expression distribution, providing provable privacy guarantees and resisting expression recognition and neural network attacks.