JVET-Q0117: Parameter sets cleanup
- Change DPS to DCI
- Move SPS id to the front of SPS
Merge request reports
Activity
-
It seems, this MR contains all three adopted aspects of JVET-Q0117?
- JVET-Q0117: Add constraint to require that all DPSs have the same content. Decision (expression of existing intent): Adopt.
- JVET-Q0117: In the PTL syntax structures in the DPS, the value of dps_max_sublayer_minus1 will be replaced with 0, to avoid signaling sub-layer levels in the DPS. Decision (BF): Adopt (unless further discussion requested with text to clarify what the semantics would be if we don’t do this). -JVET-Q0117: Add DPS constraint “When a DPS is present, each OLS in a CVS in the bitstream shall conform to at least one of the PTL syntax structures in the DPS” and remove the first sentence in the second paragraph of 7.4.4.1 General profile tier, and level semantics: “When the profile_tier_level( ) syntax structure is included in a DPS, the OlsInScope is OLS that includes all layers in the the entire bitstream that refers to the DPS.” Decision (expression of existing intent): Adopt.
- JVET-Q0117/Q0045: Remove dps_id from DPS conditioned on review of the text. Also consider renaming DPS. Decision (expression of existing intent): Adopt.
Do we have a decoder check that verifies that all DCI have the same content?
added 1 commit
- f6b75b4a - Romve some "DPS" related instructions missed.
-
We should have the conformance checks. It even seems seems, existing checks are removed by this code.
The parameter set manager contains code that can check whether the new parameter set is identical to the active one. I think, this could still be used although the thing is not called a parameter set anymore.
Also it seems the check for the NAL unit type constraint flags was removed?
To my understanding, DCI is now more like a SEI, just for some supplementary information. DPS_id is removed, so there is no ID for a DCI and DCI is not referred by any other syntax structures. And DCI does not influence the decoding process. Actually, it looks that there is no definition of "active" for a DCI in the spec. In such a case, it seems that the conformance checks are not necessary in the code, and this is also Ye-Kui's opinion. If the conformance check is implemented, what I can do is to record the first decoded DCI and compare it with the following DCI. If they are different, the decoder will be terminated. But since DCI does not affect the decoding process, such a termination seems to be an over-killing. Please let me know your opinion. If the check is mandate, I can add it.
