N0473 – Deblocking of Internal Transform Boundaries

[bheng]

The following code will cause the deblocking process to stop if a CU is not aligned with the 8x8 filtering grid. However, internal transform boundaries can still fall on the 8x8 filtering grid even if the CU does not. So, I believe the following lines should be deleted.

void LoopFilter::xDeblockCU( CodingUnit& cu, const DeblockEdgeDir edgeDir )
{
  ...

  if (edgeDir == EDGE_HOR)
  {
    if (!((cu.block(COMPONENT_Y).y % 8) == 0))
      return;
  }
  else
  {
    if (!((cu.block(COMPONENT_Y).x % 8) == 0))
      return;
  }

  ...
}

[ksuehring]

MR 515 was submitted related to this issue.

​https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM/merge_requests/515

[AnandMeher]

Dear Brian, Experts,

My understanding is that the current design of deblocking filter still does not filter any of the "sub-pu" edges and "sub-tu" edges if the corresponding CU edge does not align with the 8 x 8 sample grid.
I believe that some of the CE tests under CE5-2 category also address this problem.

Thank You

Best Regards
Anand Meher Kotra

[Kenneth]

Dear Anand, Brian, Experts,

Anand is correct that there is a CE on deblocking on 4x4 grid that takes care of this issue and also takes care of deblocking of corresponding CU boundaries on 4x4 grid. I suggest we wait with considering this until next meeting. We then also hopefully knows if we decide to go for 4x4 grid deblocking.

Best regards
Kenneth

[kiranmisra]

It would be nice if there is a test in the CE where deblocking does not skip CU's not on the 8x8 luma grid