JVET-AF0122 AhG10: Lagrange multiplier optimization for chroma ALF and CCALF [S.-W. Xie, Y. Gao, M.-H. Jia, Y.-X. Bai, C. Huang, P. Wu (ZTE)]

This contribution reports further results for the RDO Optimization method in ALF, which was proposed in JVET-AD0136 and JVET-AE0123. Regarding the proposed method, the chroma Lagrange multiplier will be adjusted by a scaling factor when the luma ALF APS transmission flag is ON and the number of history chroma ALF APS is insufficient. With VTM-22.0 and ECM-10.0 as anchors and test platforms, the reported overall PSNR-Y, U, V BD-rate and {EncT, DecT} variations of this proposal are as follows:

On top of VTM-22.0:

• LDB: 0.05%, -0.80%, -1.18% {100%, 100%}
• LDP: 0.02%, -0.89%, -1.03% {100%, 101%}

On top of ECM-10.0:

• LDB: 0.04%, -1.89%, -1.60% {100.0%, 100.7%}
• LDP: 0.11%, -1.60%, -1.45% {100.2%, 100.3%}

During the discussion of JVET-AE0123, it was suggested that the multiplier might need to be adjusted for ECM compared to VTM. The proponent reported that experimentations had been conducted to adjust the multiplier, and results show that using the same value for VTM and ECM achieves the most balanced performance among luma and chroma components.

Cross checker reported that VTM results have been completed, but ECM results are still partial at the time of presesntation. VTM results and partial ECM results are matched with that provided by the proponent.

It was asked why this has no effect on RA config. This could be because for RA config parallel encoding is performed.

Whereas the performance for VTM is considered to be overall beneficial (the chroma gain seems to justify the small loss in luma), the ECM performance is less clear. This is also the first time results had been seen for the ECM. Further tuning might provide more balanced performane for ECM.

Several experts supported including this into the VTM, but keeping it turned off for CTC.

Decision (SW): Adopt JVET-AF0122 to the next release of VTM, turned off by default.