Opened 3 months ago

Closed 3 months ago

#418 closed defect (fixed)

Mismatch with spec on LfnstDcOnly derivation and lfnst_idx parsing

Reported by: forayr Owned by:
Priority: minor Milestone: VTM-6.0
Component: VTM Version: VTM-5.2
Keywords: Cc: ksuehring, XiangLi, fbossen, jvet@…

Description

This concerns JVET_O0472_LFNST_SIGNALLING_LAST_SCAN_POS integration.

The parsing of lfnst_idx is conditioned to the value of LfnstDcOnly (called skipLfnst in vtm).
In the spec, LfnstDcOnly = 1 indicate that for all transform unit of the transform tree (which are not transform skipped), the last subblock is DC coded.

In the vtm, the array lastScanPos[] used to derive skipLfnst is updated for each transform unit, therefore it contains the value of lastScanPos for the last transform unit for each component.
So skipLfnst indicate that for each component the last transform unit which is not transform skipped in the transform tree is DC coded.

To sum up, the spec consider the last subblock for all transform unit, whereas the vtm consider the last transform unit.

Change history (2)

comment:1 Changed 3 months ago by jason

Thank you for the report.

Considering interaction with JVET-O0213 (LFNST up to 64x64), there is no mismatch between the master branch and the current spec.

However, when a macro for JVET-O0213 is disabled and intra coding is allowed for 128x128 CU (non-CTC case), the mismatch you mentioned occurs.

The fix for this issue was submitted as
https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM/merge_requests/795

No BD-rate changes under the CTC.

comment:2 Changed 3 months ago by XiangLi

  • Milestone set to VTM-6.0
  • Resolution set to fixed
  • Status changed from new to closed

Fixed as suggested.

Note: See TracTickets for help on using tickets.