International organisation for standardisation organisation internationale de normalisation

5.10Block structures and partitioning

5.10.1RQT

5.10.1.1.1.1.1.1.1JCTVC-H0315 Non-CE2: Separate RQT structure for Y, U and V components [J. An, X. Zhao, X. Guo, S. Lei (MediaTek)]

This contribution proposes a method using separate residue quad tree (RQT) structure for Y, U and V components. In HM5.0, the splitting transform unit (TU) information, signalled at TU level, is shared by Y, U and V components. In the proposed method, the splitting information of Y, U and V is signalled separately. Moreover, the maximum RQT depth values indicated in the sequence parameter set (SPS) are also separated for luma and chroma. It was reported that, by setting the proposed maximum chroma RQT depth to one, 1%~5% BD bit rate reduction was achieved for chroma. It was also reported that more BD bit rate reduction was achieved for chroma by setting the maximum chroma RQT depth greater than one. The run time was reported to be similar to HM5.0.

The suggestion in this proposal was to define separate split flags for Y, Cb, Cr.

Gain of 1-3% for chroma BD BR were reported, sometimes higher.

Increased encoder complexity was noted. No action was taken on this.

5.10.1.1.1.1.1.1.2JCTVC-H0542 Non-CE2: Cross-verification of Separate RQT for Y,U and V from MediaTek (JCTVC-H0315) [L. Guo (Qualcomm)] [late]

5.10.1.1.1.1.1.1.3JCTVC-H0123 RQT: Merge transform_tree and transform_coeff [Y. Shibahara, T. Nishi (Panasonic)]

This contribution proposed to merge the transform tree syntax and transform coefficient syntax for simplification. In WD5.0, split_transform_flag and coded_block_flag are coded in transform_tree, and transformed coefficients are separately coded in transform_coeff even though the structure of these two trees is reportedly identical. An experimental result reportedly shows negligible differences in coding efficiency (in the range of 0.001%) with no increase in software runtime relative to HM5.0 for all configurations in the common test conditions.

This was a request to attach encoded transform coefficients to the transform tree. It proposed a reordering of the bitstream, with no change in bit rate.

This would allow to remove significant redundancy from the WD description.

One of the main reasons to have this the way it is was the existence of two entropy coders with different coding of transform coefficients.

It was asled where the chroma cbf is sent, and whether there would be implications when the chroma tree has a different depth (as considered in CE2).

It was commented that there could be a problem with the position of delta-QP signalling.

It was remarked that this change was desirable, although it requires a substantial amount of changes.

Decision: Adopt

Clarification is needed regarding whether this interferes with potential adoption of CE2 (different chroma RQT depth) – the proponent of H0123 was asked to work with CE2 proponents to clarify the situation, although it was agreed that adoption of H0123 has a higher priority than CE2 (both for software and draft text, and that a unified tree must be the basis for any CE2-related subsequent modification).

5.10.1.1.1.1.1.1.4JCTVC-H0128 Cross-Check of H0123 on merge transform tree and transform coefficient [A. Saxena, E. Alshina, F. Fernandes (Samsung)] [late]