Future Video Coding Technologies: A Performance Evaluation of AV1, JEM, VP9, and HM

Abstract

This paper presents and discusses an objective performance evaluation of two lines of development for the next generation of video coding schemes: AV1 of the Alliance for Open Media (AOM) and JEM of the Joint Video Exploration Team (JVET). Their already established predecessors, as given by VP9 for AV1 and High Efficiency Video Coding (HEVC) for JEM, serve as references for this evaluation process. A large testset of 28 video sequences with different content, various resolution, and different frame rate has been used as the common data basis. For each sequence and each test candidate, four different rate points associated with varying qualities of reconstruction have been generated. When evaluated in this way relative to their corresponding predecessors, AV1 and JEM show a quite significant coding-efficiency improvement in terms of averaged Bjøntegaard-Delta (BD) bit rate of about 23% and about 32%, respectively. Moreover, this performance study also reveals that there exist different trade-offs between compression efficiency and computational complexity in terms of encoder run time. While the JEM encoder requires a factor of about 8.5 in run time relative to HM, the corresponding run time factor for the AV1 encoder is about 55.8 relative to VP9. When comparing the next-generation coding schemes among each other, this study reveals that AV1 requires an average BD bit-rate overhead of about 49% at the same objective quality while consuming 2.47 times the run time of the JEM encoder.