Semantics and perceptual coding for cloud-edge collaborative game streaming

Abstract

In cloud–edge collaborative game streaming, maintaining high perceptual quality under constrained bandwidth is challenging. To address this, we propose a perceptually-driven enhancement coding framework that enhances visual quality without increasing bitrate. Specifically, the enhancement represents the pixel-wise difference between full-quality and base-quality frames generated on the cloud server, which is transmitted to the client to refine the locally rendered baseline visuals. Our method introduces three key strategies: game-semantics-aware QP Control to dynamically guide QP distribution based on gameplay context, saliency and JND-guided QP allocation, and Non-ROI high-frequency suppression using variable rate shading with perceptually guided filtering to reduce irrelevant visual details before enhancement computation.