LiFteR - Unleashing Learned Codecs in Video Streaming with Loose Frame Referencing

Explore a groundbreaking conference talk from NSDI '24 that introduces LiFteR, an innovative video streaming system designed to enhance the performance of learned codecs. Delve into the challenges faced by deep neural network-based codecs in video streaming and discover how LiFteR's novel loose frame referencing (LFR) paradigm overcomes these limitations. Learn about the three key components of LiFteR: the LFR video dispatcher, LFR learned codec, and streaming supports for adaptive bitrate streaming. Examine the impressive performance gains achieved by LiFteR, including up to 23.8% QoE improvement over existing learned systems and a 3.2× frame rate boost. Gain insights into the future of video streaming technology and its potential impact on quality of experience for users.