On the Evaluation of Coarse Grained Parallelism in AV1 Video Coding

Abstract

Video compression is arguably one of the most time consuming tasks, especially at high resolutions. For this reason parallelization has been used at various levels: from coarse grained schemes that consider groups of blocks or frames as granules to fine grained approaches that parallelize the compression within a block of pels, e.g., at motion estimation. Given that the decade old H.264/AVC is showing its age in the 4K era, AV1 was recently launched by AOMedia as a royalty free standard promising significant compression gains (for the same quality) over the older H.264/AVC and VP9 protocols. Since AV1 is currently at its infancy, most initial research efforts concentrated on identifying possible strengths and shortcomings of the standard at various settings, focusing on video quality and compression efficiency. However, time performance and parallelism potential are equally important, particularly for media providers operating in the Cloud. In this paper we provide evaluation results of coarse grained parallelism methods implemented by the reference AV1 codec.