JVET-F0089 AHG7: Comments on metrics for extended colour volume content [E. François, C. Chevance, F. Hiron (Technicolor), D. Rusanovskyy, A. K. Ramasubramonian (Qualcomm)] [late]
This contribution discussed the metrics specified in the JEM Common Test Conditions on HDR (JVET-E1020) for evaluating coding performance on extended colour volume materials and addressed in AHG7. It reported observations made on the metrics currently considered in the AHG7, and made several suggestions on the potential usage of the wPSNR metric.
The proposed wPSNR performs weighting based on the same criterion that is used for QP adaptation in the HDR anchor. Therefore, it basically gives a preference to algorithms which do it similarly. This is not desirable.
The wPSNR is not symmetric because the original and decoded may have different luminance.
A BoG (coordinated by A. Segall) was asked to work on the following:
- to further discuss metrics for HDR quality assessment
- to coordinate the preparation of the CfE
- to review the available new content
Exploration experiments, joint meetings, BoG reports, and summary of actions taken
Exploration experiments
The setup of Exploration Experiments was discussed, and an initial draft of the EE document was reviewed in the plenary (chaired by JRO). This included the list of all tools that are intended to be investigated in EEs during the subsequent meeting cycle:
- EE1: Intra prediction (continuation)
- JVET-F0033, JVET-F0054, JVET-F0055, JVET-F0104
- The 10 aspects listed in the EE document are the minimum set expected; if additional information is brought about better combinations of the tools investigated in the EE, such combinations should be considered as well, but add-on elements not known so far shall be considered like new contributions.
- EE2: Decoder Side Motion Vector Derivation (continuation)
- EE3: Adaptive QP for 360° video
- JVET-F0038, JVET-F0049, JVET-F0072
- It was further suggested to add the case of rotated projection, which is known to improve coding efficiency.
EE4: 360° Projection Modifications and Padding (HM only)
Li Zhang was asked to compile the EE document with remote assistance by Elena Alshina, to be circulated by Thursday and reviewed Friday.
It was agreed to give the editors the discretion to finalize the document during the two weeks after the meeting, and to circulate/discuss it on the reflector appropriately.
Joint meetings
Two joint meeting sessions were held with the parent bodies.
Joint meeting Monday 1600–1800 on 360° video and tile sets (JCT-VC, JVET, VCEG, MPEG R & S)
(Also reported in the JCT-VC report.)
Liaison statements
- From SG 16 to MPEG (m40692), with remarks on
- Approved texts
- Preliminary CfE issuance
- Requirements
- From MPEG to SG 16 TD 61/Gen
- Approved texts
- Prelim CfE
360° video
- Coordination on mathematical conventions JCTVC-AA0033
- The OMAF WG 11 N 16636 has the x axis pointing to the right, z pointing back, and y pointing up
- In the draft SEI message, pitch (θ) should be airplane tilt up, yaw (ϕ) is anti-clockwise rotation (as viewed from the origin), and roll is anti-clockwise rotation (as viewed from the origin)
- Software alignment may be difficult, but we will try to proceed on that basis
- For cube map projection JCTVC-AA0034 / JCTVC-AA0035 (& arrangement of the faces) – the software has a default, which seems like a good candidate for selection – but no action was taken on this yet, although it reportedly has some desirable properties
- Widely supported in VR applications
- Removes extreme warping of the poles – more “sensible”, spatially uniform coding format
- Suitable for tiled streaming
- Memory bandwidth consumption reduction for rendering
- Coding efficiency results vary from sequence to sequence, with more reported maximum benefit than maximum loss in some test, but seams cause visual artefacts, so it was suggests that we cannot make a firm conclusion
- For assignment of future type codes:
- OMAF has written criteria currently in N 16639 (industry support, visual quality benefit, complexity), which could be the basis of establishing a plan
- Evaluation of projection mappings was discussed – JVET was continuing to work on a subjective testing methodology, and planned to have some method documented as an output of this meeting
- Decision: Angular syntax values should use 32 bits in increments of 2−16.
- Geometry type and projection type are unused JCTVC-AA0022 – remove geometry type, no action at this time on projection type
- Rectilinear viewport (equations like face of cube map, essentially indicating the position and orientation of an ordinary camera) JCTVC-AA0035. The bitstream would have an ordinary view, but with an indication of the location it represents in a spherical space – For further study
- Director’s suggested view JCTVC-AA0036 – (since updated with ID values – Decision: OK)
- Errata issues identified – corrective action planned
- MCTS extraction with slice reordering JCTVC-AA0029 – For further study
- MCTS extraction of non-neighbouring regions JCTVC-AA0024 #11 (item withdrawn)
- MCTS hash JCTVC-AA0028 – No action.
- Maximum number of MCTS tile sets JCTVC-AA0024 (note: already standardized) – consider some fix of the extraction SEI message
Joint meeting Wednesday 1600–1730 on CfE preparation (JVET, VCEG, MPEG R & S)
In preliminary perceptual testing, in some cases, the JEM with up to 35% lower data rate performed visually better than HEVC HM.
An updated draft of the CfE was reviewed. One aspect was to request bitstreams to be submitted by 6 July. Additionally, JEM 6.0 anchors were planned to be provided by June 10:
- Samsung and Qualcomm were asked to provide anchors for SDR and perform mutual cross-check.
- Samsung was asked to provide the anchors for 360°, and InterDigital to cross-check them. Furthermore, HM anchors for 360° need modification of adaptive search range (InterDigital).
- Sharp, Qualcomm and Technicolor were asked to provide HM and JEM anchors for HDR, and to mutually perform crosschecks.
Small changes were made in the test conditions. An SDR bit rate point was modified. A different part of one sequence was used. Some bit rates were adjusted for HDR. For 360° video, one test sequence was added and some biy rate points were changed.