GRACE: Loss-Resilient Real-Time Video through Neural Codecs

Explore a groundbreaking conference talk on GRACE, a loss-resilient real-time video system that utilizes neural codecs to maintain video quality across varying packet loss scenarios. Delve into the challenges of real-time video communication and learn how GRACE addresses them through joint training of neural encoders and decoders. Discover how this innovative approach outperforms conventional loss-resilient schemes, significantly reducing undecodable frames and stall duration while improving video quality. Examine the extensive evaluation process, including various videos, real network traces, and a comprehensive user study involving 240 participants. Gain insights into the future of real-time video communication and its potential to enhance user experience in high-latency network environments.