JVET-O0205 CE8-related: Improvements on BDPCM [S. Yoo, J. Choi, J. Nam, J. Heo, J. Lim, S. Kim (LGE)]

Two methods for improving BDPCM are proposed in this contribution as follows.

• Intra mode derivation for BDPCM
• Line-shaped CG for BDPCM block

In the current VVC, when the BDPCM mode is enabled for a block, its prediction mode is set to MODE_INTRA and its intra mode is inferred to be the first candidate in the MPM list, which is planar mode. Considering that BDPCM actually uses one of intra prediction between horizontal and vertical modes, this proposal suggests to derive the BDPCM prediction mode based on actual prediction mode. Another aspect is to use line-shaped CG for BDPCM. In the BDPCM process, once the DPCM direction is indicated to a CU, the transform skipped residuals are processed/predicted by the line. Therefore, combining the residuals with a line CG seems to be desirable and it also provides coding performance. As a result, coding gains of the proposed scheme is observed to be -0.18% / -0.28% in Class F and TGM in AI, -0.12% / 0.00% in Class F and TGM in RA, and -0.24% / -0.23% in Class F and TGM in LDB, respectively.

First aspect: See under JVET-O0315

Second aspect: Results are presented only for combination with the first aspect. Though it gives some additional gain for AI in TGM, the gain that the first method gives in RA is going to zero in the combination. Furthermore, decoder run time increases. No action.