Back to Search Document details
16th Meeting: Geneva, October 2019 2019-09-24 20:16
Non-CE3: Enable MIP prediction for 64×N or N×64 blocks at maximum transform size 32
Authors: Zixiang Zhang Ericsson K. Andersson Ericsson D. Saffar Ericsson Rickard Sjöberg Ericsson J. Ström Ericsson Ruoyang Yu Ericsson
Abstract
In the current VVC, the maximum transform size is configurable to be equal to 32 or 64. The MIP prediction is able to be used by a coding block when a) the block width ∕ height ratio and height ∕ width ratio is equal or less than 4; b) the block width is equal or less than the maximum transform size and the block height is equal or less than the maximum transform size. This means that when the maximum transform size is configured to be equal to 32, a coding block which has a dimension 64×64, 64×32, 64×16, 32×64 or 16×64 is not be able to be coded as a MIP predicted block.
Meeting Notes
JVET-P0198 Non-CE3: Enable MIP prediction for 64×N or N×64 blocks at maximum transform size 32 [Z. Zhang, K. Andersson, D. Saffar, R. Sjöberg, J. Ström, R. Yu (Ericsson)]

In the current VVC, the maximum transform size is configurable to be equal to 32 or 64. The MIP prediction is able to be used by a coding block when a) the block width ∕ height ratio and height ∕ width ratio is equal or less than 4; b) the block width is equal or less than the maximum transform size and the block height is equal or less than the maximum transform size. This means that when the maximum transform size is configured to be equal to 32, a coding block which has a dimension 64×64, 64×32, 64×16, 32×64 or 16×64 is not be able to be coded as a MIP predicted block.

This contribution proposes to enable the MIP prediction for 64×N and N×64 coding blocks when the maximum transform size is configured to be equal to 32, where N ranges from 16 to 64.

The simulation (non-CTC) uses VTM6.0 as the anchor software and the maximum transform size is configured to be equal to 32 (Log2MaxTbSize=5).

The proposed method has a luma BDR impact for AI/RA of -0.02%/-0.03% compared to VTM6.0 with Log2MaxTbSize=5 and decoding time impact for AI/RA of 100%/100% compared to VTM6.0 with Log2MaxTbSize=5.

See the notes for the joint proposal JVET-P0803

References:
JVET-O0545
Cited:
JVET-P0831 X JVET-P0834 X JVET-P0969
ARCHIVE v2 :
JVET-O2001-vE_proposal_text_JVET-P0198-v1.docx JVET-P0198-v2.docx JVET-P0198-VTM-6.0-Log2MaxTbSize_5_psnr.xlsm JVET-P0198-VTM-6.0-Log2MaxTbSize_5_psnr_4_decimal.xlsm
Decisions
See the notes for the joint proposal JVET-P0803
Citation