YuZu - Neural-Enhanced Volumetric Video Streaming

Explore a groundbreaking approach to volumetric video streaming in this USENIX conference talk. Dive into the world of YuZu, a neural-enhanced system that drastically improves the quality of 3D immersive viewing experiences. Learn how the researchers achieved a 542× speedup on super resolution inference, developed a Quality of Experience model, and integrated these components into a holistic system. Discover how YuZu outperforms existing viewport-adaptive solutions, boosting QoE by up to 178%. Gain insights into the technical aspects of intra- and inter-frame optimizations, resource adaptation, and line-rate adaptive super resolution for volumetric video streaming.

Intro
Volumetric Video
Our Approach: Yuzu
YuZu Overview Server
Intra-frame SR
SR Performance Breakdown
Yuzu's QoE Improvement
YuZu's End-to-end Performance
YuZu vs. Viewport-Adaptive Streaming
Demo