Back to Search Document details
16th Meeting: Geneva, October 2019 2019-10-02 17:44
[AHG18][Non-CE5] Proposed cleanup of deblocking filter process
Authors: Shunsuke Iwamura NHK Shimpei Nemoto NHK Atsuro Ichigaya NHK K. Andersson Ericsson Ruoyang Yu Ericsson J. Enhorn Ericsson
Abstract
This contribution proposes cleanup of deblocking filter process which was originally discussed on 3rd of September on the JVET reflector as a part of AHG18 activities. Two aspects related to boundary strength derivation and one aspect related to derivation process of and tC’ are proposed to align the deblocking filter process with other coding tools.
Meeting Notes
JVET-P0586 [AHG18][Non-CE5] Proposed cleanup of deblocking filter process [S. Iwamura, S. Nemoto, A. Ichigaya (NHK), K. Andersson, R. Yu, J. Enhorn (Ericsson)]

Aspect 1 of the proposal has been presented.

If at least one of the blocks have non-zero coefficient and not TS, then BS is set equal to 1. See the discussion below

//-------------------------------------

Current deblocking behaviour:

  • If both blocks are BDPCM, Bs = 0; if one side is BDPCM, it is treated as intra, Bs = 2.
  • If one blocks is palette, we do skip on the palette side.
  • TS is treated as a normal (transform) block.

Three related proposals

  • JVET-P0586. If at least one of the blocks have non-zero coefficient and not TS, then BS is set equal to 1.
  • JVET-P0571 Method 2. It is proposed to bypass the deblocking on the BDPCM side in the same way how it is done for Palette. The claim is mostly consistency between palette and BDPCM modes.
  • JVET-P0611. If one of the blocks is BDPCM, TS, or Palette, then BS = 0.

It was mentioned that BDPCM and TS do not apply transforms in a similar way, so deblocking should be applied in a similar way. It was mentioned that it would be desirable to treat three modes similarly.

It was mentioned that BDPCM is not allowed in CTC on classes E-A and is only used in the screen content.

BoG recommendation: further study outside the CE, possibly in the context of former AHG11.

//-------------------------------------

Aspect 2 has been considered in the previous discussion on …

Aspect 3.

When VTM uses scaling list, the QP values are still derived without taking it into account that may results in wrong tc and beta values. An offset is derived based on the 6*log2(m/16), where m is the representative value of the scaling list (in this case the DC). The logarithm is approximated with the LUT of size 23 bytes.

It was commented that the proposal makes sense in general but some evidence needs to determine if the action is needed.

BoG recommendation: Further study (possibly in former AHG15). In particular, it needs to be decided what quantization matrices need to be used.

References:
JVET-N1010 JVET-O0656 JVET-O2001
Cited:
JVET-P0018 JVET-P0700 X JVET-P0783 X JVET-P0815 X JVET-P2021 JVET-Q0096
ARCHIVE v2 :
JVET-P0586-BCWdb-vs-VTM6.0.xlsm JVET-P0586-declockCleanup-v2.docx JVET-P0586-HDR_BCW_DB-vs-VTM-6.0.xlsm JVET-P0586-TSdb-vs-VTM6.0.xlsm
Decisions
BoG recommendation: Further study (possibly in former AHG15). In particular, it needs to be decided what quantization matrices need to be used.
Citation