JVET-R0069 AHG12: Control of loop filtering across subpicture/tile/slice boundaries [L. Zhang, Y.-K. Wang, K. Zhang (Bytedance), Hendry, N. Park, H. Jang, J. Nam, S. H. Kim, J. Lim (LG Electronics)]

This contribution proposes to confirm the two conditionally agreed items, i.e., the following:

1. Remove the PPS flag loop_filter_across_slices_enabled_flag.
2. Move loop_filter_across_tiles_enabled_flag from PPS to SPS. Note that this flag is not signalled for each subpicture.

Thirdly, this contribution also proposes to close a topic that was left open for determination at this meeting. It is proposed that the loop_filter_across_tiles_enabled_flag moved from PPS to SPS is renamed to be loop_filter_across_boundaries_within_subpics_enabled_flag, and it is used to control whether loop filters are applied across tile and slice boundaries inside subpictures (not to control filtering across tile and slice boundaries that are also subpicture boundaries).

In the latest VVC draft text, in-loop filtering across subpicture, tile, and slice boundaries is controlled by the following syntax elements:

• loop_filter_across_subpic_enabled_flag[ i ]: for controlling of deblocking, SAO, and ALF across subpicture boundaries, signalled in the SPS, one for each subpicture.
• loop_filter_across_tiles_enabled_flag: for controlling of deblocking, SAO, and ALF across tile boundaries, signalled in the PPS, just one (thus applicable to all tiles in all pictures referring to the PPS).
• loop_filter_across_slices_enabled_flag: for controlling of deblocking, SAO, and ALF across slice boundaries, signalled in the PPS, just one (thus applicable to all slices in all pictures referring to the PPS).

One participant said that rectangular slices can be used generally rather than tiles, and argued against both aspects 1 and 2.

For aspect #1, the plan to remove the (pps_)loop_filter_across_slices_enabled_flag had been because it was considered unnecessary since we have subpictures as an alternative. It was discussed whether we want to assume that encoders would use subpictures, and was suggested not to remove this encoder flexibility. See also JVET-R0109, which requested to retain this flag.

It was agreed to not confirm the prior planned action for aspect #1 due to this concern.

The second aspect had initially been agreed to be recommended in an AHG pre-meeting, but it was agreed to overturn that recommendation.

For aspect #2, it is noted that the tile partitioning is established at the picture level, so it was commented that it would be strange to put the control over the boundaries only at a higher level.

It was thus agreed to not confirm the prior planned action for aspect #2 for this reason.

Contributions JVET-R0069, JVET-R0109, JVET-R0197, JVET-R0247 are all related.

There was discussion to consider changing the semantics for the existing flags.

The meaning of the various flags ultimately is (in the output text of the previous meeting) that whenever something indicates that filtering across a tile/slice/subpicture/virtual boundary is off, it is off, regardless of whether it may appear to be turned on by something else (except that ALF and SAO can be off for one side of a subpicture boundary and on for the other side of the boundary).

Concerns were expressed about the third aspect as well, as it would change the concept of all flags having “veto power”. So no action was taken on that aspect either.