Massively Accelerating VVC Intra Encoding Through Decoder-side Neural Quality Enhancement
Wei Geng, Xin Wang, Tianjing Zhang · 2023
The latest Versatile Video Coding (VVC) standard has offered the state-of-the-art compression efficiency but at a great sacrifice of the complexity, in which the quad-tree nested multi-type tree (QTMT) based coding unit (CU) decision is one major contributing feature. This paper proposes to massively accelerate the VVC intra encoding by only allowing the quad-tree blocks, and later compensate the performance loss through decoder-side neural quality enhancement (NQE). Having decoder-side NQE is feasible due to the availability of advanced Graphical Processing Unit (GPU) or Neural Processing Unit (NPU) in mainstream devices. The proposed approach reduces the complexity of VVC intra encoding by almost a factor of 10 with only 1.81% BD-Rate (Bjøntegaard Delta Rate) increase, against the VVC anchor.