JVET-P0486 CE8-related: Alignment of maximum transform-skip size with maximum transform block size [M. G. Sarwer, R. L. Liao, J. Luo, Y. Ye (Alibaba)]

Discussed in context of CE6 (Fri 4 Oct. 1930hrs)

Regarding the signalling of the maximum transform skip size, it is asserted that the current signalling in VVC draft 6 allows too much flexibility, which is not only unnecessary but also burdens hardware verification and conformance testing. Therefore, this contribution proposes to either remove it and set it to the same value as the maximum TB size, or to move it from PPS to SPS and to signal just as an SPS flag indicating either 32 or 64 in the same way as the maximum TB size signalling.

In VVC draft 6, the maximum transform skip (TS) block size and maximum BDPCM block size are 32x32, whereas the maximum transform block (TB) size is 64x64. This contribution proposes to align the maximum TS and BDPCM block size with maximum TB size. In the proposed method, the maximum TS block size can be up to 64x64. For larger TS and BDPCM blocks (i.e. size > 32), it is proposed to split them into 32x32 residual unit (RU). This way, the existing VVC6 TS residual coefficient scanning method can be reused, and a decoder never has to scan residual blocks larger than 32x32. It is reported that the BD-rate impact of the proposed method is -0.02% (AI), -0.13% (RA), -0.20% (LB) for the SCC sequences under CTC test conditions.

In terms of compression, no significant benefit.

As transform skip is targeting other content than regular transform, it is desirable to configure it separately. Encoders might also want to restrict its usage to small blocks only for lower complexity.

It is asserted that the current flexibility of TS is desirable.

Would be too early to decide at this meeting, needs further study. It is pointed out that TS is also studied in the context of the lossless CE, which will anyway include some study of signalling.