JVET-H0060 AHG4: SDR anchor generation for the Draft Joint Call for Proposals [H.-C. Chuang, J. Chen, M. Karczewicz (Qualcomm), K. Choi, E. Alshina (Samsung)]

In this document, information regarding the SDR anchor generation for the Draft Joint Call for Proposals is described. On the selected five UHD sequences (FoodMarket4, CatRobot1, DaylightRoad2, ParkRunning3, CampfireParty2, namely Class SDR-UHD1) and five HD sequences (BQTerrace, RitualDance, MarketPlace, BasketballDrive, Cactus, namely Class SDR-HD1), encoding results and comparison of coding performance between HM-16.16 and JEM-7.0 anchors are reported. The matched bit rates of all sequences are less than 1% from the target bit rate, and the averaged absolute deviation from the target bit rate was 0.2% and 0.0% using Constraint Set 1 and Constraint Set 2, respectively.

This was discussed Thursday morning at 11:15.

It was clarified that the anchors should not exceed target bit rates (same as condition for proposals).

It was clarified to use JEM7 anchors in the CfP, such that they can be made available to prospective proponents as early as possible.

Excel sheets for the CfP were prepared by Jianle Chen (SDR), Yuwen He (for 360°, represented by Philippe Hanhart) and Edouard François (for HDR). Side activity was conducted to unify the three Excel sheets and make them available as input document.

Configuration files shall also be made available for all test cases.

Subsequently, the CfP was discussed.

It was further decided that no additional test cases for objective measurements will be defined in the CfP. This was anyway only planned for SDR, where the CfP already has a sufficient variety of test material.

In a follow-up discussion (in the context of presenting JVET-H0030) it was decided to add another, higher rate point for HM, such that it would be possible to compute reasonable BD-rate numbers. See further notes in the section discussing JVET-H0030.

It was further clarified that this does not intend to change the current CTC, and new CTC will have to be established after the CfP anyway.

For 360, Philippe Hanhart prepared viewing sessions. Rate points for Landing and Balboa need to be confirmed.

New sequences were proposed as replacements for Harbour and KiteFlite (with moving cameras) in JVET-H0022. However, encodings only exist for the CTC; these should not be used in the CfP.

A report on anchor generation for 360° video will be provided. It was verbally reported that some of the anchors exceed the target rates, which need to be updated.

If possible, all 6 sequences should be kept in the CFP (see further notes on this under JVET-H0022).

One expert suggested to include MountainBay2 (which was in Draft CfP as a placeholder for objective testing). However, this would mean that it needs to replace one other sequence. Further, though some subjective viewing was done in the last meeting, rate points for this sequence need further clarification.

Further discussion and update was performed Sat. 21^st 1130–1300, as well as in the context of presenting the BoG report JVET-H0095. Various text improvements were made without controversial discussion. The reporting template for the SDR test cases was also presented.

It was decided that the templates should compute two comparisons (against the HM and against the JEM). This has the advantage that another anchor exists that is closer to the expected proposal performance, such that mutual comparison of BD performance differences between proposals becomes more reasonable.

Comparison of encoding/decoding time is only mandatory with the HM as reference, and is optional for the JEM.

A template had been made available as input document JVET-H0100, and the final version (3 sheets for SDR, HDR, 360) is to be included with CfP. It was later reported that the HDR and 360° versions are making progress and no specific problems were foreseen.