JVET-AN0249 Suggestions for a potential upcoming CfP [J. Samuelsson-Allendes, S. Deshpande (Sharp)]

This input contribution discusses a potential upcoming Call for Proposals (CfP) for a video coding standard with capabilities beyond VVC. Three items are discussed and advocated for in the proposal:

1. Not include general preprocessing and postprocessing in the scope and in CfP responses
2. Include compression-level dependent preprocessing and postprocessing in the scope and in CfP responses
3. Make CfP subjective test results non-anonymous with possibility to opt-out

The following proposal are made with the intention of addressing the three items above:

1. In the coding conditions text of the CfP:
   1. Disallow (or strongly discourage) the use of preprocessing and postprocessing that does not depend on the level of compression (e.g., QP)
   2. Explicitly allow (or encourage) use of preprocessing and postprocessing that depends on the level of compression (e.g., QP)
2. In the subjective test:
   1. Ask the question “How close is the processed video to the original?” rather than “What is the level of the visual quality of the video?”
3. For submissions to the CfP:
   1. Provide submitters with results of the subjective test prior to the meeting in which all results are anonymised except their own submission and the anchors.
   2. Provide two options for submitters:
      1. Have their results removed from the subjective test and not published
      2. Have their results published with the subjective test in a non-anonymous manner

The following text is proposed for aspect 1 (pre/post processing):

• Use of preprocessing and/or postprocessing that does not depend on the level of compression (e.g., QP) is discouraged.
• Use of preprocessing and/or postprocessing that depends on the level of compression (e.g., QP) is encouraged.

The first bullet is stronger than the following bullet from CfE

• Use of preprocessing and/or postprocessing shall be described.

The second bullet may be practically covered by the following bullet in CfE:

• Sequences are expected to be encoded at full input resolution for all rate points. If coding at reduced resolution is part of the algorithmic concept, it shall be described.

It was commented on the first bullet that, in a DCR test, methods that would like to improve the appearance of the output, would be deviating from the original.

It was commented on the second bullet that it is not difficult to make anything QP dependent.

It was suggested to encourage, if pre or post processing is applied, also to report results where the same is applied to the anchor.

This could be extended to some sentence of the body of the text:

“If any methods are applied in preprocessing, postprocessing or encoder optimization, it is encouraged to provide also results when applying equivalent methods to the anchor. Such information may also be requested during evaluation at the meeting to be provided.”

It was agreed to include such sentence, but otherwise keep the formulations at this moment as they were in the CfE.

It was suggested to check the formulation on low delay conditions, if it is clear enough that the sentence “Overall structural delay shall not be larger than that of the respective anchor” shall also include pre and post processing.

Aspect 2 (subjective test) is obsolete as DCR (impairment) will be used.

Aspect 3 (making results anonymous optionally):

A comment was made that it might be better to keep the current practice. Results would not be included in public report, if not desired by proponents, and anonymous documentation in intermediate presentations in JVET is hardly interpretable by people outside JVET.

A comment was made by the test coordinator that the option of not including results in the initial presentation may be difficult to implement.

It was commented that the “binary” decision might enforce unnecessary decisions to opt out.

Several other experts also expressed that the current practice should be retained.

No action taken on aspect 3.