JVET-R0286 AHG9: On general constraint information syntax [Y.-J. Chang, V. Seregin, Y. He, M. Coban, M. Karczewicz (Qualcomm)]
The proponent said approach 1 of aspect 1 does not need to be discussed.
For aspect 1 approach 2, JVET-R0173 proposes syntax conditions to skip syntax. This contribution proposes sensibility constraints on values without introducing such syntax conditions.
Decision (sensibility constraints): Adopt aspect 1 approach 2. (The editor may remove redundant expression, if any.)
For aspect 2, it was noted that JVET-R0191 also proposes such an LMCS flag and JVET-R0207 proposes such an LFNST flag.
Decision (general constraints on tools): Adopt constraint flags per aspect 2.
Aspect 3 proposes four flags relating to layered coding: single_layer_constraint_flag, single_sublayer_constraint_flag, single_sublayer_per_layer_constraint_flag, and no_inter_layer_pred_constraint_flag.
Both the first and second flags would have only one layer, so the name of the second flag seems misleading.
It was suggested that the second flag is just a combination of the first and third.
It was noted that all of this information can be found in the VPS and making this change might require VPS and SPS rewriting for the BEAM case and that the controlled features are not really coding tools per se, but rather syntax structures and functionalities.
Some of these can be equivalently expressed by specifying that the SPS VPS ID shall be equal to 0 or that the number of layers shall be equal to 1 or that vps_all_indpendent_layers_flag shall be equal to 1.
It was said that no_inter_layer_pred_constraint_flag would be equivalent to an existing VPS flag vps_all_indpendent_layers_flag.
Decision (general constraints cleanup): Adopt the single_layer_constraint_flag and no_inter_layer_pred_constraint_flag (perhaps renamed to all_layers_independent_constraint_flag).
Aspect 4 proposes to add general constraint flags for having no VPS present and always having the PH in the SH. The second of these had been agreed previously at this meeting.
For a single-layer bitstream, the VPS is purely metadata and as agreed above, we will have a way to indicate a single-layer bitstream. Prohibiting VPS presence in this case would be sort of like prohibiting the presence of an SEI message, so a constraint flag to express this did not seem necessary. So no action was taken on the VPS part of aspect 4.