Misalignment between JVET-O2001vE and VTM-6.1 on chroma deltaQP when local dual tree is used

[zhangkai]

Problem:
In the current spec, the signaling of chroma delta-QP may be controlled by the luma partitioning depth instead of the chroma partitioning depth when the local dual tree split is applied with SCIPU.
For example, a 16×8 luma block is split by TT into three blocks with dimensions 4×8, 8×8 and 4×8. Suppose the CTU size is 128. The split depth denoted as cbSubdiv in JVET-O2001-v14 is then increased from 7 to 9. Local dual tree is activated in this case and a corresponding 8×4 chroma block is intra-coded without split. In the current design, cbSubdiv is also set to be 9 for the chroma block as the luma component. Therefore, if cu_chroma_qp_offset_subdiv is equal to 7, chroma delta-QP cannot be signaled for this 8×4 chroma block. However, chroma delta-QP can be signaled for an 8×4 chroma block when it is split by single-tree without local dual-tree or regular dual-tree.
But in VTM-6.1, the signaling of chroam delta-QP only depends on the split of the chroma component.

Recommended solution: Change to spec to align with the reference SW. Attached please find a recommended WD to be changed.

[zhangkai]

working draft fixed

[delagrangep]

Agreed (as original proponent of cbSubdiv stuff).

[bbross]

Thanks for reporting and confirming.

This will be fixed JVET-P2001-vD.