Back to Search
Document details
CE2.2.1 and CE2.2.2: ALF coefficient coding and range constraints
Abstract
In the contribution, CE2.2.1 and CE2.2.2 tests for simplifying adaptive loop filter (ALF) are reported. In CE2.2.1, ALF coefficients are mapped from signed values to unsigned values and coded with 0-th order exp-Golomb coding. It is reported that compared with VTM2.0.1 (using adaptive k-th order exp-Golomb coding), 0-th order exp-Golomb coding for ALF coefficients results in 0.08%, 0.14%, and 0.12% luma BD-rates for AI, RA, and LB, respectively, and very minor run time changes. In CE2.2.2, the ranges of ALF coefficients are limited to achieve 10-bit representation. For the non-center ALF coefficients, the number of required bits to represent a coefficient is reduced from 11 to 10 by constraining the real value range within [-1.0, 1.0). For the center ALF coefficient, the number of required bits to represent a coefficient is reduced from 15 bit to 10 bit by constraining the real value range within [0.0, 2.0). It is reported that compared with VTM2.0.1, the reduction of ALF coefficient ranges achieves 0.00%, 0.00%, and -0.01% luma BD-rates for AI, RA, and LB, respectively, and almost no run time changes. It is claimed that reducing ALF coefficient ranges is effective for reducing hardware cost.
JVET-L0082 CE2.2.1 and CE2.2.2: ALF coefficient coding and range constraints [Y.-C. Su, C.-Y. Chen, Y.-W. Huang, S.-M. Lei (MediaTek)]
References:
Decisions
JVET-L0082 CE2.2.1 and CE2.2.2: ALF coefficient coding and range constraints [Y.-C. Su, C.-Y. Chen, Y.-W. Huang, S.-M. Lei (MediaTek)]
Citation