Back to Search Document details
16th Meeting: Geneva, October 2019 2019-09-25 03:17
On Scaling for Transform Skip
Authors: T. Tsukuba Sony M. Ikeda Sony Y. Yagasaki Sony Teruhiko Suzuki Sony Takeshi Tsukuba
Abstract
VVC has two scaling processes in inverse transform skip as “tsShift” and “bdShift”, which are jointly conducted as “iTransformShift (= bdShift - tsShift = maxLog2DynamicRange - bitDepth - (log2(nTbW) + log2(nTbH))/2)” in SW. However, considering concept of transform skip, there should be no scaling process in transform skip. This contribution proposes to remove scaling process in transform skip by combining scaling (tsShift, bdShift in 8.7.2) for transform skip into scaling (bdShift in 8.7.3) in inverse quantization process.
Meeting Notes
JVET-P0272 On Scaling of Transform Skip [T. Tsukuba, M. Ikeda, Y. Yagasaki, T. Suzuki (Sony)]

VVC has two scaling processes in inverse transform skip as “tsShift” and “bdShift”, which are jointly conducted as “iTransformShift (= bdShift – tsShift = maxLog2DynamicRange – bitDepth – (log2(nTbW) + log2(nTbH))/2)” in SW. However, considering concept of transform skip, there should be no scaling process in transform skip. This contribution proposes to remove scaling process in transform skip by combining scaling (tsShift, bdShift in 8.7.2) for transform skip into scaling (bdShift in 8.7.3) in inverse quantization process.

It is reported that:

For ClassA1/A2/B/C/D/E sequences, average BD-rate differences are 0.00%, 0.00% and 0.00% for AI, RA and LB, respectively.

For ClassF sequences, average BD-rate differences are 0.00%, 0.00% and 0.00% for AI, RA and LB, respectively.

For SCC sequences, average BD-rate differences are 0.00%, 0.00% and -0.04% for AI, RA and LB, respectively.

No separate presentation – see the notes for JVET-P0538.

References:
JVET-N1010 JVET-O2001
Cited:
JVET-P0809 X JVET-P1000
ARCHIVE v1 :
JVET-P0272.docx JVET-P0272_CTC.xlsm
Decisions
No separate presentation – see the notes for JVET-P0538.
Citation