Opened 2 years ago

Closed 2 years ago

#135 closed defect (invalid)

DCT2 transform matrix down-sampling might be wrong in the spec

Reported by: zhou Owned by:
Priority: minor Milestone:
Component: spec Version: VVC D3 v8
Keywords: Cc: ksuehring, bbross, XiangLi, fbossen, jvet@…

Description

The matrix down-sampling is only done for columns, it should be done for rows too.

Replace

If trType is equal to 0, the following transform matrix multiplication applies:

y[i]= ∑_("j" =0)("nonZeroS" -1)▒〖transMatrix[i][j*2(6-Log2(nTbS)) ]*x[j]〗 with i = 0..nTbS − 1 (8 567)

with

If trType is equal to 0, the following transform matrix multiplication applies:

y[i]= ∑_("j" =0)("nonZeroS" -1)▒〖transMatrix[i(6-Log2(nTbS))][j*2(6-Log2(nTbS)) ]*x[j]〗 with i = 0..nTbS − 1 (8 567)

Change history (6)

comment:1 Changed 2 years ago by zhou

  • Summary changed from DCT2 transform matrix down-sampling might is wrong in the spec to DCT2 transform matrix down-sampling might be wrong in the spec

comment:2 Changed 2 years ago by bbross

  • Version set to VVC D3 v6

comment:3 Changed 2 years ago by bbross

  • Version changed from VVC D3 v6 to VVC D3 v7

comment:4 Changed 2 years ago by bbross

The downsampling is done the same way as it is done for HEVC, so is there a bug in HEVC as well?
(see equation (8-319) of HEVC v5)

comment:5 Changed 2 years ago by bbross

  • Version changed from VVC D3 v7 to VVC D3 v8

comment:6 Changed 2 years ago by bbross

  • Resolution set to invalid
  • Status changed from new to closed

As already mentioned it is the same as in HEVC where the first nTbS basis functions are selected from the Matrix (no subsampling!) and then subsampled (this is the column subsampling).
See also Fig. 6.2 in the HEVC book.
So no issue here.

Note: See TracTickets for help on using tickets.