JVET-R0174 LFNST index signalling [C. Rosewarne, J. Gan (Canon)]
With the adoption of JVET-Q0784 LFNST is only able to be applied in the luma channel for slices using a shared coding tree. When applied, the chroma TBs must use DCt-2. In WD8 and VTM-8.0 the signalling of lfnst_idx is conditioned on the last significant position of the TBs in the CU, i.e. luma and chroma TBs for shared coding trees. Since the chroma TBs are using DCT-2, there is no need to check their last position for the purpose of suppressing lfnst_idx. This contribution proposes to restrict last significant coefficient checking for lfnst_idx suppression to the luma channel only for CUs in shared coding trees. BD-rate impact under RA is 0.01%, 0.01%, 0.03%, no effect in AI (dual tree used instead) or LDB (LFNST disabled).
It is mentioned that the condition that is suggested to be removed by this proposal was introduced by purpose to avoid buffering issues.
No action was taken on this.