JVET-R0342 AHG9: A Summary of Proposals Related to HRD [S. Deshpande (Sharp)]

This contribution intends to provide a summary of proposals on core aspects of HRD (including HRD operation, related SEI message signalling and sub-bitstream extraction).

The following proposals are covered in this summary: JVET-R0094, JVET-R0100, JVET-R0101, JVET-R0103, JVET-R0264, JVET-R0295, JVET-R0297, JVET-R0413.

It is suggested that this summary be used for the reviewing of these proposals, such that the discussions may be done in a more structured and efficient manner.

Related to HRD Signalling and operation:

1. Signal a fixed DPB output time offset for each temporal sublayer, controlled by a presence flag, within the buffering period SEI message and use these offsets to calculate picDpbOutputDelta[ i ]? (JVET-R0094)

See the notes for that contribution.

1. Not signal and infer dui_sublayer_delays_present_flag[ bp_max_sublayers_minus1 ] to be 1, to make sure du_spt_cpb_removal_delay_increment[ bp_max_sublayers_minus1 ] which is used for inference of other syntax is always signalled in DUI SEI message? (JVET-R0100 Proposal 1)

Decision (sensibility cleanup): Adopt this aspect.

1. Signal the for loop for syntax elements dui_sublayer_delays_present_flag[ i ] and du_spt_cpb_removal_delay_increment[ i ] in the reverse order or signal du_spt_cpb_removal_delay_increment[ bp_max_sublayers_minus1 ] before other du_spt_cpb_removal_delay_increment[ i ] for i < bp_max_sublayers_minus1? (JVET-R0100 Proposal 2)

No need for action was identified for this aspect.

1. Add missing inference rules for the alternative timing information related syntax elements cpb_alt_initial_cpb_removal_delay_delta[ i ][ j ], cpb_alt_initial_cpb_removal_offset_delta[ i ][ j ] which are used for HRD operation? (JVET-R0101 Proposal 1)

Decision (expression of existing intent): Adopt this aspect.

1. Fix asserted bugs related to unspecified or missing length of u(v) coded syntax elements by referring to correct length syntax elements? (JVET-R0101 Proposal 2)

Decision (editorial bug fix): Adopt this aspect.

1. Not signal du_common_cpb_removal_delay_flag, du_common_cpb_removal_delay_increment_minus1[ i ], num_nalus_in_du_minus1[ i ] and du_cpb_removal_delay_increment_minus1[ i ][ j ] in picture timing SEI message when there is only one DU in an AU and condition the duCpbRemovalDelayInc variable derivation? (JVET-R0103 Proposal 1)

Decision (sensibility cleanup): Adopt this aspect.

1. Perform the editorial fixes for the CPB operation section to define and use AU and DU related variable names correctly? (JVET-R0103 Proposal 2)

Decision (editorial bug fixes): Adopt.

1. Apply an asserted editorial clarification of what n0…n5 mean for the number of conformance tests in Annex C? (JVET-R0297)

Decision (editorial bug fixes): Adopt.

1. Fix an asserted bug for the inference rule for syntax elements in the syntax structure sublayer_hrd_parameters( ) to use syntax from SPS or VPS as appropriate instead of always using sps_max_sublayers_minus1? (JVET-R0297)

Decision (editorial bug fixes): Adopt.

1. Fix an asserted bug for n3 specifying number of conformance test for IRAPs that are not CRAs with associated RASL pictures and alternative timing? (JVET-R0297)

This was requested be refined in offline work to account for GDR. The contribution was revised to account for this. This was further discussed on 24 April at 0630 UTC (chaired by GJS).

Decision (expression of existing intent): Adopt.

1. Fix an asserted bug in the derivation of InitialCpbRemovalDelay in C.2.3 (Timing of DU removal and decoding of DU) aligning it with derivation in C.2.2 (Timing of DU arrival) to account for alternative timing? (JVET-R0297)

Decision (editorial bug fixes): Adopt.

1. Fix an asserted bug by adding missing text to update values of CpbDelayOffset and DpbDelayOffset in C.2.3 -Timing of DU removal and decoding of DU? (JVET-R0297)

Decision (editorial bug fixes): Adopt.

1. Fix an asserted bug in equation C.11 for derivation of NominalRemovalTime[ n ], to account for CpbDelayOffset? (JVET-R0297)

Decision (editorial bug fixes): Adopt.

1. Signal a separate set of alternative buffering delay parameters for VCL HRD and for NAL HRD and use them for HRD operation? (JVET-R0413)

There had been a previous contribution JVET-Q0219 proposal #1 that discussed this, and further study had been conducted.

Decision (bug fix): Adopt.

Related to sub-bitstream extraction:

1. Apply following quoted modifications to sub-bitstream extraction process? (JVET-R0264)

“When (pps_)mixed_nalu_types_in_pic_flag is equal to 0”, remove from outBitstream all NAL units for which all of the following conditions are true:

– nal_unit_type is not equal to IDR_W_RADL, IDR_N_LP, or CRA_NUT.

– nuh_layer_id is equal to LayerIdInOls[ targetOlsIdx ][ j ] for a value of j in the range of 0 to NumLayersInOls[ targetOlsIdx ] − 1 inclusive, “but not equal to OutputLayerIdInOls[ targetOlsIdx ][ j ] for a value of j in the range of 0 to NumOutputLayersInOls[ targetOlsIdx ] − 1 inclusive”.

– TemporalId is greater than or equal to NumSubLayersInLayerInOLS[ targetOlsIdx ][ j ].

This is intended as a correction for expression of existing intent. It was agreed that something needs to be done for the mixed NAL unit type issue, but not as proposed. After offline study, the contribution was revised to account for this. This was further discussed on 24 April at 0635 UTC (chaired by GJS).

After the offline study, a participant commented that adding a use of the PPS content in the sub-bitstream extraction process (rather than just using NAL unit types) seemed undesirable. Another option would be to add a constraint that when max_tid_il_ref_pic_plus1 is equal to 0 (which means that only the IRAP pictures in the reference layer would be allowed to be used for reference) it would be prohibited to have mixtures of IRAP and non-IRAP NAL units in the picture.

Decision (bug fix): Add a constraint that when max_tid_il_ref_pic_plus1 is equal to 0 (which means that only the IRAP pictures in the reference layer would be allowed to be used for reference) it would be prohibited to have mixtures of IRAP and non-IRAP NAL units in the picture.

1. Use the bit rate indicated in general_hrd_parameters( ) and ols_hrd_parameters( ) of the OLS for derivation of the subpicture bit rate variables SubpicBitRateVcl and SubpicBitRateNal in the subpicture level information SEI message in Section D.7.2? (JVET-R0295)

There had been an equivalent proposal at the previous meeting JVET-Q0395.

The proponent said that the method that had been incorporated at the previous meeting could potentially result in CPB overflow. Another participant asked if the modification could produce underflow. The proponent said that the scheme follows what the encoder has expressed in the syntax.

It was commented that some of the discussion at the previous meeting may have been before acting on subpicture HRD modelling as currently specified, and that now that we have this, the proposal makes sense.

Decision (bug fix): Adopt.

1. Apply asserted simplification changes to the subpicture extraction process Annex C.7 to constrain subpictures in multi-layer bitstreams to be aligned in terms of boundary position and subpicture ID across layers? (JVET-R0295)

It was discussed whether the alignment of boundaries that is assumed in the proposal is fundamentally part of our current design, and it was noted that a contribution JVET-R0058 had some relationship to this. However, it was commented that the proposal is compatible with contribution JVET-R0058.

This was further discussed in the closing plenary on 24 April at 1440 (chaired by GJS & JRO) after the discussion of JVET-R0058.

Prior to the adoption of JVET-R0058, the draft did not allow the subpicture IDs to be different across different layers. In that context, the proposal is purely editorial, as basing the extraction process on the subpicture IDs provides the same extraction.

Editor action item: The editor is asked to consider rephrasing the subpicture extraction process to use subpicture IDs or indexes in order to simplify the description. If the editorial expression is simpler after accounting for the JVET-R0058 aspect, it is suggested to use this expression.

1. Fix an asserted bug which rewrites the cbr_flag[ tIdTarget ][ j ] to 1 for all CPBs when sli_cbr_constraint_flag is equal to 1 in subpicture extraction process? (JVET-R0295)

Decision (expression of existing intent): Adopt this aspect.

1. Also perform several assertedly minor fixes (e.g. #903) and editorial improvements in Annex C.6 (Sub-bitstream extraction process) and C.7 (subpicture sub-bitstream extraction process)? (JVET-R0295)

Decision (editorial bug fixes): Adopt this aspect.

Discussion stopped here for JVET Track A on 19 April at 0700 UTC.