JVET-Q0125 CE5-related: JC-CCALF with Alternative Filter Weight [C.-W. Kuo (Panasonic)]
This contribution was further reviewed in Track B Monday 13 January at1330 (chaired by JRO).
In this contribution, it is proposed to use individual JC-CCALF filter weight for each alternative filter. Specifically, instead of using various weight values as in JC-CCALF, only sign changes are applied for 4 alternative filters, e.g., + − − +. In decoder, the current CTU selects the filter weight according to the parsed alternative filter index. Compared to the JC-CCALF in CE5-3.2, a coding gain improvement and negligible runtime change are reported.
Simulation results with joint YUV performance (AHG13 8:1:1 metric) are summarized as follows.
All Intra | Random Access | Low Delay B | ||||||||||
Y | U | V | YUV | Y | U | V | YUV | Y | U | V | YUV | |
CCALF (CE5 anchor) | 0.17% | -8.31% | -7.06% | -1.40% | 0.14% | -10.14% | -9.82% | -1.88% | 0.12% | -13.00% | -9.48% | -2.15% |
JC-CCALF (CE5-3.2) | 0.10% | -5.84% | -4.87% | -0.99% | 0.16% | -8.63% | -8.27% | -1.56% | 0.12% | -12.71% | -9.47% | -2.12% |
Alternative filter weight | 0.10% | -6.65% | -5.56% | -1.14% | 0.15% | -9.39% | -9.16% | -1.74% | 0.11% | -13.03% | -10.11% | -2.22% |
The gain comes from two aspects: Replacing the weight by +/−1 (sign), the gain comes mainly from saving bits for signalling. In total, there are less options for weighting, but more options for the joint filter. The adopted option from CE5-2.1 still has more options for the filters, and the new proposal still is less well performing. The complexity of the adopted proposal is not a real concern, and the approach proposed here appears to have some more irregularity.
No action was taken on this.