range of ph_cu_qp_delta_subdiv_intra/inter_slice

[leesh007]

According to the current spec, the range of ph_cu_qp_delta_subidv_intra/inter_slice is defined as
2 * ( CtbLog2SizeY − MinQtLog2SizeIntra(Inter)Y + ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma ).

However, it seems not correct.

I belive that it should be modifed as

2 * ( CtbLog2SizeY − MinQtLog2SizeIntra(Inter)Y) + ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma.

[yk]

Can experts in this area confirm? Thanks!

[leesh007]

Dear Experts,

Accoding to my understandings, "cbSubdiv" value increases by 2 in case of QT or TT and increases by 1 in case of BT.

However, mttDepth increases only 1 both BT and TT.
Consequently, "ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma" value has range of 0 to 2*( CtbLog2SizeY − MinCbLog2SizeY ) in the spec.

Therefore, I beleive that multiplying by 2 for "ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma" is not reasonable for the range decision of "ph_cu_qp_delta_subidv_intra/inter_slice".

[jiechen]

I think the current spec is correct.
The maximum value of cbSubdiv should be "2 * ( CtbLog2SizeY − MinQtLog2SizeIntra(Inter)Y + ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma )" rather than "2 * ( CtbLog2SizeY − MinQtLog2SizeIntra(Inter)Y) + ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma".

As you said, cbSubdiv is increased by 2 in case of TT, but mttDepth is only increased by 1 in TT split. So "ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma" should also be multiplied by 2 if we consider TT, otherwise the range is not sufficient for cbSubdiv to reach the minCb in same cases.

BTW, "2*" in the maximum value of "ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma" is for BT case. So they are two separated issues.

[leesh007]

Dear Experts,

In case of small TT partition with cbSubdiv +2, it is impossible to split their child-nodes to reach mttDepth equal to ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma
becauuse its width or height should not be smaller than MinCbSize.

To be specific, if all parents nodes for leaf node are coming from small TT region,
its max mttDepth should be just half of ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma.

Thus I'm still curious whether ph_cu_qp_delta_subidv_intra/inter_slice needs to have such a large ranges.

[bbross]

Thanks for bringing that up and I am trying to understand the issue. Can the current range
a) cause illegal splits?
b) allow just signalling unnecessary large values but no illegal splits are introduced
c) be fully exploited as in the current spec (no issue here)

[jiechen]

Replying to bbross:

Thanks for bringing that up and I am trying to understand the issue. Can the current range
a) cause illegal splits?
b) allow just signalling unnecessary large values but no illegal splits are introduced
c) be fully exploited as in the current spec (no issue here)

The is the range of the QG size. The currenbt range allows just signalling unnecessary large values but no illegal splits will not be introduced. Similar situation also exists in the range of "ph_max_mtt_hierarchy_depth_intra(inter)_slice_luma"

But the suggested range here is not sufficient. If we change the range to the suggested one, then it will loss ability to make the QG size as small as CU size in some cases. It violates our design intent.

[bbross]

Nothing broken so action taken as agreed in the meeting.