JVET-O0200 CE8-related: RDPCM intra mode mapping, maximum RDPCM block size [M. Coban, Y. Han, M. Karczewicz (Qualcomm)]

This contribution proposes to assign horizontal/vertical RDPCM directions to horizontal/vertical intra prediction modes instead of planar intra mode. The simulations results show Class F: -0.12% AI, -0.09% RA, 0.02% LB BD-Rate, and Class TGM: -0.18% AI, -0.18% RA, -0.04% LB BD-Rate, versus the VTM-5.0 anchor under common test conditions. The maximum coding block width/height that supports RDPCM is aligned to maximum transform skip size signalled in PPS instead of fixed 32 maximum dimension.

Other contributions that have the same aspect of Part 1 (hor/vert direction): 205, 296, 315, 342, 463, 542. See further notes under JVET-O0315.

Part 2 suggests to inherit the maximum size of BDPCM (hard coded with 32) with the maximum size of TS (signalled). The rationale behind limiting the size of BDPCM was motivated by a throughput issue, which is no longer existing since BDPCM is applied in the coefficient domain.

In the current syntax, TS is always inferred when BDPCM is enabled. Therefore, coupling the two max sizes is logical.

JVET-O0206 also proposes this aspect, but raises another issue that there would be an ill-posed condition if TS was disabled and BDPCM was enabled. It therefore suggests another condition at low level to resolve this issue.

During the discussion it is agreed to better resolve this at a high level, defining that the BDPCM flag can only be set if TS is set. As the BDPCM flag is agreed to be introduced in SPS (as per adoption of JVET-O0165), the appropriate solution would be to also move the TS flag to SPS (suggested in JVET-O0783 option 2). The PPS flag should be removed. Max TS size should stay at PPS

Decision: Adopt JVET-O0200 “Part 2” (align max size of BDPCM and TS)

Proponents of JVET-O0165, JVET-O0200, JVET-O0783 were asked to work on unified syntax of those aspects.