JVET-AL0052 AhG17: Comments on proposed draft text: Call for Evidence on video compression with capability beyond existing standards [E. Alshina, T. Solovyev (Huawei)]

This document suggests testing procedures for potential Call for Evidence on compression with capability beyond existing standards. The major focus is to address requirements of providing coding solution efficient in constrained encoder configuration.

Difficulties in designing constrained encoder test conditions:

• It is impossible to realistically emulate ‘real time’ encoding during standard development
• Encoder constrains are highly dependent on platform, use case, device.
• Standard development shall not become a competition of product level optimized encoders
• Run time cross-check for both encoder and decoder will be required

Cross-check for both encoder and decoder.

• Decoder executables are submitted as always
• Encoder executables are submitted as part of responses in ‘constrained encoder category’
• Encoder cross-check is slow? Check just few cases (for example, the lowest and the highest rate for several selected sequences)
• Useful to submit executables generated with same settings /optimization level for anchor

Different variants are presented, where variant 4 is most flexible, and also uses the same VTM anchor (only one) as in the “compression benefit” testing

It was commented that for the fast encoding capability it might not be desirable to impose decoder constraints. It would also not be desirable to make submissions of too many complexity points mandatory, and to impose too many restrictions in general. It is valuable for the CfE if additional information about the capability of fast encoding is provided with a submission. One purpose is also to possibly improve conditions for a subsequent CfP, and learn which might be reasonable constraints to impose there.

It was commented that it would be desirable to have multiple VTM configurations, if possible including some which go down to HM speed. Another expert suggested to investigate a range which goes up to 4.5x encoding runtime of VTM for submissions. It should however be observed that this might not be working with current rate 1, and more effort to generate anchors.

Diverging opinions on whether also decoder complexity should be limited.

It should be clear that in a CfE not such an amount of detail can be achieved in assessment as in a rigid standards development.

It was agreed that the approach of variant 4 is a reasonable step, and that towards the next meeting a “dry run” should be prepared, using VTM anchor(s) and configurations of ECM/NNVC etc. to exercise the methodology. To be further worked out in a BoG (E. Alshina, F. Bossen) with mandate to

• identify the test cases to be prepared for the next meeting in the testing for fast encoding basede on VTM, ECM and NNVC
• discuss the open questions above
• review contributions related to testing conditions targeting reduction of run time.
• after meeting with parent bodies, and if the plan to execute such a test is agreed, to discuss a first draft of section 4 in JVET-AL0047