Fast CU Depth Decision Algorithm for AVS3

The third generation of Audio Video Coding Standard (AVS3) achieves more than 20% coding performance improvement compared with the previous video coding standards, HEVC and AVS2. AVS3 adopted a more flexible CU partition scheme, which brings promising coding efficiency but makes the computational complexity much higher than HEVC and AVS2. In this paper, we propose fast algorithms for CU partitioning based on the CU depth. The spatial neighboring depth information is used to predict a QT depth range of current CU, according to which the partition tree can be befittingly pruned. On this basis, we predict a proposed BET (BT and EQT) depth to further skip some partitions. The threshold to skip partition processes is decided from statistics. The experimental results show that the combination of the proposed methods can achieve 22.07% time-saving with 0.57% BDBR loss under Random Access configuration and 45.17% complexity reduction with 0.50% BDBR loss under All Intra configuration.