Work

Fine-Grained Loss Recovery via Packet-Level Content-Awareness for Ultra-low-latency Video Communications

Year: 2026
Type: conference-paper
Abstract: Ultra-low-latency video applications, such as cloud gaming and VR, increasingly deliver high-bitrate frames within tens of milliseconds. As each frame may contain hundreds of packets, uniformly retransmitting every lost packet can unnecessarily block an entire frame, even when the missing packet has little rendered-quality impact. This paper presents LetItGo , a receiver-side content-aware loss-recovery mechanism based on the observation that lost packets within the same frame can have very different impacts on frame quality. LetItGo uses packet position and pre-decoding to decide whether a lost packet is worth waiting for. It retransmits high-impact losses and lets the existing decoder conceal less critical ones, enabling fine-grained recovery without changing the sender, encoder, playback-decoder interface, or transport protocol. We implement LetItGo in WebRTC for H.264 and extend it to H.265. Experiments with high-resolution videos and real cloud-gaming traces show that LetItGo reduces transmission time by 9.41% on average over a strong loss-recovery baseline while keeping rendered quality within a 0.44 dB-SSIM gap. Results across H.264 and H.265 at both 30 and 60 fps show that the benefit is not tied to a single codec.
Language: English
Open Access status: gold
OpenAlex ID: W7227856176
FWCI:
Cites: 56
Cited by:
Related to: 10