Back to Search Document details
12th Meeting: Macao, October 2018 2018-10-08 07:29
CE4-related: Generalized bi-prediction improvements combined from JVET-L0197 and JVET-L0296
Abstract
This contribution presents several methods to improve BMS2.1 generalized bi-prediction (GBi), including coding unit (CU) size constraint for GBi, early termination of affine motion estimation for GBi weights conditionally, pre-selection of GBi weight pairs for low-delay pictures in the encoding for 1-pel and 4-pel MVD precision, conditionally skipping bi-prediction search when reference pictures of two lists are the same, skipping RD cost checking for unequal GBi weights based on temporal layer and the POC distance between reference picture and current picture, and changing floating calculation to fixed-point calculation for unequal GBi weights. Moreover one bug fix related to 6-parameter affine motion estimation in GBi is also included. Compared to BMS2.1 with VTM configuration (GBi off), simulation results show that GBi with the proposed improvements and CU size constraint that disables GBi for CU size smaller than 256 luma samples achieves -0.66% BD-rate with 6% encoding time increases and no decoding time increases for RA.
JVET-L0646 CE4-related: Generalized bi-prediction improvements combined from JVET-L0197 and JVET-L0296 [Y.-C. Su, C.-Y. Chen, Y.-W. Huang, S.-M. Lei (MediaTek), Y. He, J. Luo, X. Xiu, Y. Ye (InterDigital)] [late]
Decisions
JVET-L0646 CE4-related: Generalized bi-prediction improvements combined from JVET-L0197 and JVET-L0296 [Y.-C. Su, C.-Y. Chen, Y.-W. Huang, S.-M. Lei (MediaTek), Y. He, J. Luo, X. Xiu, Y. Ye (InterDigital)] [late]
Citation