New DRS framework boosts live streaming by 9% BD-rate reduction

Conventional adaptive bitrate (ABR) streaming relies on static bitrate ladders, which ignore content-specific characteristics and often waste bandwidth. Per-title optimization improves efficiency but requires pre-encoding analysis that is impractical for live streaming. To solve this, researchers propose Dynamic Resolution Switching (DRS), a framework that dynamically selects optimal resolutions at each bitrate using real-time analysis. DRS augments static ladders with strategically chosen representations guided by user bandwidth distributions and cross-over prediction.

The core of DRS is a lightweight, bitstream-based video quality metric (VQM) trained on Pairwise Comparison datasets to maximize cross-over accuracy. At each bitrate, the VQM evaluates all candidate representations to pick the resolution that maximizes subjective quality. This decision can be made per segment, allowing fine-grained adaptation. Experimental results show a significant 9% BD-rate reduction under the proposed VQM, all while maintaining compatibility with existing streaming protocols. The paper was accepted at IEEE ICIP 2026.