JVET-R0053 AHG9: Signalling tile partitioning [S.-T. Hsiang, C.-M. Tsai, Y.-W. Huang, S.-M. Lei (MediaTek)]
Items 1 and 2 of this contribution belong in this category.
This was discussed in AHG Session 1.2 Monday 6 April at 1600 UTC (chaired by GJS & YKW).
This contribution proposes three high-level syntax modifications related to signalling tile partitioning of the coded picture, summarized as follows:
- The number of explicitly provided tile column widths or tile row heights is proposed to be allowed to be equal to 0. When no_pic_partition_flag is equal to 0 and the coded picture contains only one tile, the proposed method can infer the tile column width and the tile row height to be equal to the picture width and the picture height, respectively, without signalling the syntax elements tile_column_width_minus1[ 0 ] and tile_row_height_minus1[ 0 ] for deriving the tile column width and the tile row height.
JVET-R0285 and JVET-R0080 are said to contain the same proposal. It is a syntax optimization to avoid signalling tile width or height if the picture is not split into tiles in the corresponding dimension. It was commented that this is consistent with another use of avoiding explicit signalling. It was commented that for uniform tile size signalling, this would increase the signalling that is needed. No action on this was recommended; although it saves a few bits in the PPS in some cases, it adds more in another case that some participants expect to be more common.
- pps_log2_ctu_size_minus5 is proposed to be signalled in the picture parameter set (PPS) only when rect_slice_flag is equal to 0, single_slice_per_subpic_flag is equal to 0 and num_slices_in_pic_minus1 is greater than 0.
A participant commented that it is nice to know pps_log2_ctu_size_minus5, even in other conditions where it is not necessary for parsing. Since it is only two bits, it does not seem important to avoid sending this. No action was recommended on this.
- (pps_)loop_filter_across_tiles_enabled_flag is proposed to be signalled only when the number of the tiles in the coded picture is greater than 1.
At the previous meeting, it was planned that we would move this flag to the SPS (see notes for JVET-Q0120).
JVET-R0113 and an aspect of JVET-R0197 are proposing this in principle as well.
It was asked whether bitstream merging would be affected by this, and it was commented that PPSs need to be rewritten in that case anyway.
The initial AHG Recommendation was to confirm the move to the SPS, but this was overturned as noted under JVET-R0069.
This was further discussed on 22 April at 1715 (chaired by GJS & YKW).
Decision (cleanup): Adopt conditioning presence of loop_filter_across_tiles_enabled_flag per JVET-R0113 aspect 1 (avoiding redundant signalling without change of functionality).
JVET-R0053 AHG9: Signalling tile partitioning [S.-T. Hsiang, C.-M. Tsai, Y.-W. Huang, S.-M. Lei (MediaTek)]
Item 3 of this contribution belongs to this category. See the notes in section 6.2.2.1 for JVET-R0053.