Back to Search
Document details
CE2-related: Early awareness of accessing temporal blocks in sub-block merge list construction
Abstract
In current VVC design, a sub-block merge list is constructed in a way that both ATMVP candidate and affine candidates are inserted in order. For both the ATMVP candidate and affine candidates, they rely on temporal motion information from collocated picture but may be from different locations. In addition, for some cases, the ATMVP candidate may be unavailable. This contribution proposes three modifications: 1) Only 1 spatial neighboring block, instead of 4, is checked to derive initial MV; 2) ATMVP candidate is always available; 2) remove TMVP in affine candidate construction process. With the proposed method, the decoder could immediately know whether loading temporal motion information is needed after paring the sub-block merge candidate index. In addition, the decoder is able to skip the ATMVP candidate construction process based on the decoded candidate index. Simulation results reportedly show that the proposed methods achieves 0.02% and 0.09% BD rate reduction for RA Main10 and LDB Main 10 configurations, respectively.
JVET-M0273 CE2-related: Early awareness of accessing temporal blocks in sub-block merge list construction [L. Zhang, K. Zhang, H. Liu, J. Xu, Y. Wang, P. Zhao, D. Hong (Bytedance)]
References:
Decisions
merged
JVET-M0273 CE2-related: Early awareness of accessing temporal blocks in sub-block merge list construction [L. Zhang, K. Zhang, H. Liu, J. Xu, Y. Wang, P. Zhao, D. Hong (Bytedance)]
Citation