JVET-AF0311 Results of expert viewing at AF meeting for the spatial scalability category of the VVC multilayer VT M. Wien (RWTH Aachen University)
See section 7.5 for the initial discussion in a joint session. This document reports the results of an on-site expert viewing performed at the 32nd JVET meeting in Hannover, DE.
The evaluation targets at the spatial scalability category of the test plan document JVET-AE2021, comparing the compression performance of dual layer (DL) coding with a downscaled base layer at scaling ratio 2 to single layer coding at the low resolution with upscaling to the target resolution using the VVC interpolation filters (UP) as well as single layer coding at the target resolution (SL).
19 experts participated in the expert viewing. With a comparably low number of valid scores, the results should be considered as an indication of tendencies. They are not regarded as a necessarily conclusive result.
With respect to the goals of this experiment, the following observations are made:
- Since the UP sequences represent the extracted base layer of the DL streams, the impact of the enhancement layer can be studied. It can be seen that in most cases, the MOS only slightly increases with the inclusion of the enhancement layer. This might be attributed to the fact that other effects overlay the benefit of enhanced detail provided with the DL stream.
- The result show a similar relation of the SL, DL, and UP results compared to JVET-AE0288 for the sequences CorsaireJoy. For TallBuildings2, the results seem to indicate a visual improvement of the DL configuration compared to the UP configuration. This might be attributed to the encoder modifications reported in JVET-AF0186.
- The MOS results for the DrivingPOVLogo sequence seem to indicate that there might be a range of benefit for the DL scheme at rates at or below 1Mbit/s.
- The MOS results for the NxBoxeLogo sequence seem to indicate bitrate ranges where UP, DL, and SL have benefits, similar to what might be chosen in a bitrate ladder setting.
It is recommended to take these results into account by reconsiderering the test setup with respect to bitrate ranges to be tested for certain resolutions and encoder configurations. The test sequence CorsaireJoy might not include sufficient detail to provide a suitable demonstration case for the targeted application.
An initial version had been presented during a joint meeting (see section 7.5); the final version was presented in JVET on Friday Oct. 20 1115-1140.
It is notable that the upscaled base layer has a very high quality when presented to viewers without compressing it. This may suggest the conclusion that the ranges of rates where spatially scalable coding with conveying information in the enhancement layer need to be carefully selected.
It was however commented that the conclusion that the uncoded upsampled base layer would be almost transparent might be questionable, as it had been tested along with coded sequences with visible distortions. Another test methodology would be necessary to investigate whether it is transparent compared to the uncoded full resolution.
It was commented that in some cases of low rate, dual layer coding becomes better than single layer at full resolution. It was pointed out that when using RPR in an optimized this might change (as also upsampled base layer is better in those cases). Unavailability of an optimized encoder criterion might be a problem here.
However, the purpose of this verification test is not to demonstrate superiority of multi-layer coding over single layer, but rather demonstrating that multi-layer functionality works reasonably well with VVC.
It was concluded to take the results into account in developing the new version of the multi-layer VVC test plan JVET-AF2021.
AHG4: Test material (2+1)
Contributions in this area were discussed at 1405–1440 on Wednesday 18 Oct. 2023 (chaired by JRO).
JVET-AF0311 Results of expert viewing at AF meeting for the spatial scalability category of the VVC multilayer VT M. Wien (RWTH Aachen University)
Some results of experts viewing experiments conducted during the meeting were shown.
Using the referencing settings proposed in JVET-AF0186 were reported to help.
At the low rates, an upsampled lower resolution video seemed better. [add notes from contrib]
There is some more coherence in the results than had been obtained previously, but it is not yet clear how to resolve the issues.
It was commented that the encoder still seems not mature and needs further work.
Generally it seems needed to conduct further study of the issues raised in these VVC multilayer comparisons reported in JVET-AF0311.
Session ended at 13:00.
BoGs (1)
The following break-out groups were established at this meeting to conduct discussion and develop recommendations on particular subjects.