JVET-P0418 Non-CE3: Cleanup of MRLP Line Storage [T. Hellman, M. Zhou, B. Heng (Broadcom)]
The Multi-Line Reference (MLR) Intra prediction mode in VVC requires the storage of 3 neighboring luma reference lines with a CTU to generate predictions. The Cross-Component Linear Model (CCLM) tool also requires 3 neighboring luma reference lines for its downsampling filters. This contribution proposes to modify the definition of MLR to use the same 3 lines as CCLM. It claims that this change aligns MLR with CCLM in terms of neighboring pixel referencing and reduces the storage requirements for decoders. Reported VTM-6.0 test results show a 0.01% RA coding loss from this change.
AI BD-rate is 0.04%. Benefit is one less line to be stored and number of lines is aligned with CCLM tool. At top of the CTU both tools use single line.
It was commented by non-proponent that this simplification is good to have and that the current design that skips one line is strange. The proposal cleans up and simplifies the MRL line usage.
The BoG recommended adoption.
Cross-check in JVET-P0980.