VVC Adaptive Loop Filter Optimization for Subpicture-based Viewport-adaptive Streaming

Virtual reality (VR) systems require delivering high-fidelity 360° video content to immerse viewers to the captured scene. The viewport-adaptive streaming (VAS) methods have been developed to deliver 360° VR content efficiently. The Versatile Video Coding (VVC) standard introduces the subpicture picture partitioning tool, which creates isolated regions suitable for VAS. The usage of Adaptive Loop Filter (ALF) as a VVC in-loop filtering operation is limited in subpicture-based VAS. This paper aims at enabling usage of ALF in subpicture-base VAS through proposing a set of encoding constraints that are standard compliant. While ALF is activated, the proposed constrains guarantee that no coding coordination with respect to sharing of ALF parameters among subpictures is required. This further allows subpicture-based parallel encoding of high-resolution VR content. We study the performance of several methods targeting both single- and multi-thread encoding platforms. The experimental results indicate that VR content encoding can be parallelized at a subpicture-group level, while still preserving most of the ALF gain. The proposed method with subpicture-group encoding parallelization achieves on average -2.4%, -4.0%, and -4.3% Bjøntegaard delta rate reduction for Y, U, and V components respectively, compared to the case where ALF operation is deactivated.