JVET-AP0123 Matrix based MPM prediction [Y. Gao, H. Tian (Shandong Univ.), S. Li, J. Lei, B. Li, F. Xing, P. Han (Hisense)] [late]

In EE2, the MPM list is constructed to find the best RD mode for the current block based on the best RD mode of neighbours. That is because the best RD mode of the current block is related to the best mode of neighbours. These best mode is related to each other, yet not necessarily exactly the same. Thus a matrix based MPM prediction is presented with, where the relationship between the best mode of the current block and the neighbours is explored. And no additional complexity is introduced.

On top of ECM-19.0, results of the proposed method are:

AI: -0.01% -0.07% -0.04% , EncT: % , DecT: %.

Partial results in RA show losses.

It was suggested to not use CTC sequences for training (currently, CTC sequences and BVI have been used).

Further investigation may be necessary on handling non-angular modes.

No action at this moment.