JVET-AD0100 AHG12: Block vector guided CCCM [R. G. Youvalari, D. Bugdayci Sansli, P. Astola, J. Lainema (Nokia)] [late]

This contribution proposes a new convolutional cross-component model which uses block vector of the co-located IBC or intraTMP coded luma block to identify the reference area for deriving the CCCM model. The proposed block vector guided CCCM (BVG-CCCM) method uses the same 7-tap filter design of the CCCM method and the co-located luma block’s samples for prediction. The mode is signalled with a context coded flag which is conditioned on co-located luma block’s mode. The impact on coding efficiency and runtimes over ECM-8.0 is reportedly {for Y, U, V, EncT, DecT}:

CTC classes: AI { -0.00%, -0.10%, -0.10%, 100%, 100%}, RA { -0.xx%, -0.xx%, -0.xx%, 10x%, 10x%}

Class F: AI { -0.06%, -0.51%, -0.57%, 101%, 100%}, RA { -0.xx%, -0.xx%, -0.xx%, 10x%, 10x%}

Class TGM: AI { -0.64%, -1.20%, -1.03%, 101%, 100%}, RA { -0.09%, -0.27%, -0.26%, 97%, 10x%}

Combination with EE2-1.8 shows higher gain also for natural content.

It was agreed to investigate this in an EE. A configuration using the method only for IBC should additionally be tested.