JVET-R0068 AHG8/AHG9/AHG12: Miscellaneous HLS topics [Y.-K. Wang, L. Zhang, Z. Deng, J. Xu, K. Zhang, K. Fan (Bytedance)]

Item 7 of this contribution belongs to this category. The contribution proposes to change the syntax name of (pps_)init_qp_minus26 to init_qp_minus32 and update the semantics accordingly.

In HEVC the QP range is from −QpBDoffset to 51, so 26 is approximately the midpoint for 8 bit video.

In VVC the QP range is from −QpBDoffset to 63. For 10 bit video this is −12 to 63.

It was commented that the typical QP in VVC is also a little higher than in HEVC.

It was noted that there is a 37 in the proposed semantics that should be 31.

It was commented that the same change had been proposed in L0553 and had not been adopted for the same reason as noted above.

There was some discussion of what we think the typical QP would be for use of VVC.

It was agreed not to take action on this.

JVET-R0068 AHG8/AHG9/AHG12: Miscellaneous HLS topics [Y.-K. Wang, L. Zhang, Z. Deng, J. Xu, K. Zhang, K. Fan (Bytedance)]

Item 4 of this contribution belongs to this category.

JVET-R0068 AHG8/AHG9/AHG12: Miscellaneous HLS topics [Y.-K. Wang, L. Zhang, Z. Deng, J. Xu, K. Zhang, K. Fan (Bytedance)]

Discussion started here in JVET Track A on 21 April at 1300 UTC (chaired by GJS, JRO, YKW).

Items 3, 8, and 9 of this contribution belong to this category.

Item 3 proposes to change the semantics of sps_affine_amvr_enabled_flag equal to 1 to use the wording of "may be used" instead of "is used".

Decision (editorial bug fix): Adopt (clarify that this has “one way” semantics).

Editor action item: There should be a general review of the word “may”, for wording consistency and clarity – e.g., whether it expresses permission.

See notes under JVET-R0049 which is equivalent on item 8.

Item 9 proposes to signal five_minus_max_num_affine_merge_cand instead of five_minus_max_num_subblock_merge_cand when sps_affine_enabled_flag is equal to 1, and MaxNumSubblockMergeCand is derived as a sum of the maximum allowed number of sbTMVP candidates and the maximum allowed number of affine candidates.

Only when affine is enabled is “five_minus_max_num_subblock_merge_cand” signalled. The proponent said this seemed confusing.

Another participant said that this parameter is needed only when affine is enabled, and there is not really a problem here that needs to be solved. If affine is not enabled, the number of candidates is known to be 1.

Another participant commented that this could create a situation where the number of affine merge candidates is signalled to be 0 but affine mode will be used, which seems too strange.

JVET-R0068 aspect 9, JVET-R0215, JVET-R0371, and JVET-R0373 are related.

See the notes for JVET-R0371.

JVET-R0068 AHG8/AHG9/AHG12: Miscellaneous HLS topics [Y.-K. Wang, L. Zhang, Z. Deng, J. Xu, K. Zhang, K. Fan (Bytedance)]

Discussed in AHG Session 1.5 (chaired by GJS, JRO & JB).

Item 2 of this contribution belongs to this category.

Discussed UTC Tuesday 7 April, 2020, 13:00. Chaired by JRO and JB.

The proposal is to use a special entry in the level signalling in the PTL syntax structure to indicate the property of an all-IRAP subset of the bitstream.

It was commented that the encoder could use TemporalId 0 for IRAP pictures. However, it was commented that this will make one less sublayer to be used.

It was commented that in intra based trick play, some IRAP pictures would be output/displayed multiple times.

It was commented that trick play can also use non-intra pictures, and scene cuts may appear in arbitrary positions and can be coded as IRAP pictures.

It was commented that such information, if useful, should use an SEI message.

It was commented that the information is useful, somewhat like marking non-reference pictures etc.

It was commented that DASH already makes use of this information, as a bitstream property.

It was commented that the information may be hard for the encoder to figure it out and set. However, it was counter-argued that this is similar for setting the level for temporal scalable layers.

It was asked what people think about having an SEI message for indicating that the 0-th level in the PTL is for IRAP only.

This was further discussed on 23 April at 1525 (chaired by GJS).

This uses an offset of 1 so that index 1 is for sublayer 0. There is a similar offset for max_tid_il_ref_pics_plus1[ i ][ j ].

The proposal did not include text of an extraction process to extract this extra subset, or an HRD model for such a subset.

It was commented that this is not necessarily how a decoding system would use intra-only trick play.

This was deferred for further study for a potential future SEI message.

JVET-R0068 AHG8/AHG9/AHG12: Miscellaneous HLS topics [Y.-K. Wang, L. Zhang, Z. Deng, J. Xu, K. Zhang, K. Fan (Bytedance)]

Item 6 of this contribution belongs to this category.

JVET-R0068 AHG8/AHG9/AHG12: Miscellaneous HLS topics [Y.-K. Wang, L. Zhang, Z. Deng, J. Xu, K. Zhang, K. Fan (Bytedance)]

Item 5 of this contribution belongs to this category.

JVET-R0068 AHG8/AHG9/AHG12: Miscellaneous HLS topics [Y.-K. Wang, L. Zhang, Z. Deng, J. Xu, K. Zhang, K. Fan (Bytedance)]

Item 1 of this contribution belongs to this category.

Item 1 proposes to require that slice_type shall be equal to 2 (intra slice) in the following cases (in addition to being required under some other conditions):

i) intra_only_constraint_flag is equal to 1

ii) the NAL unit type is an IRAP NAL unit type and the current picture is the first picture in the current AU.

The first case would be redundant, and we generally don’t discuss implications of constraint flags outside of their semantics.

Decision (expression of existing intent): Specify that slice_type shall be equal to 2 (intra slice) when the NAL unit type is an IRAP NAL unit type and the current picture is the first picture in the current AU.