Video @Scale Fall 2022 LIVE Q&A #3
summary
Cloud-driven metaverse delivery claims 25 Mbps minimum for full offload and up to 5–6× bandwidth savings using gaze-driven rendering, but end-to-end latency targets range from 30–120 ms and are hard to hit. Practical demos discuss server-side ML with optional lightweight client models, live object removal feasibility, RGBD or mesh transfer for partial offload, and 10× frame interpolation, upscaling and stabilization tools — concrete tradeoffs between bandwidth, latency, and where inference runs.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:12Real-time object detection
- 4:17ML model trade-offs
- 6:25Eye-tracking limits
- 8:32Low-latency gaze delivery
- 10:03Cloud video bandwidth
- 12:08Partial offloading
- 14:15AI super-sampling
- 17:03Q&A
related talks
10mRTC @Scale, Future RTC Experiences Session - Live Q&A
41m8K Codec Low-Latency Livestreaming, and AI Streaming
50mDemuxed 2023 Lightning Talks
17mYuZu: Neural-Enhanced Volumetric Video Streaming
19mLow Latency Live from a Different Vantage Point
21mLeaving the lag behind
19mBeyond the Screen: Streaming the Immersive Internet with Mawari
25mFirefly: Untethered Multi-user VR for Commodity Mobile Devices