JVET-R0484 Report of 360 verification test planning side activity [V. Baroncini, J. Boyce, J.-R. Ohm, M. Wien, Y. Ye]
This contribution was presented Wed. 22 April 1625–1650.
This is the summary report of the 360 video verification test planning side activity, Tuesday April 21 15:15 – 16:25 UTC.
It was agreed to use 5 360 video sequences in the formal verification tests.
When doing initial simulations, it was agreed to use more sequences, and plot their R-D performance to get quality and rate range behaviour of each sequence.
SDR used QP 26-46, step size of 4 for VVC, and QP 26-46, step size of 1 for HEVC
It was agreed to use the same QP settings for 360 content in the preliminary round of simulations.
It was suggested that, after preliminary encoding is done, perform preliminary viewing of the content, and suggest viewports (one static and one dynamic).
It was commented that once bitstreams are available, generating viewport images is relatively easy.
It was agreed to separate encoding and viewport selection/generation into different tasks, with possibly different volunteers.
It was commented that SDR bitstreams have two settings, one ALF on and the other with ALF off.
It was agreed that we will keep ALF on for 360 video encoding in the preliminary round of simulations.
VVC CfP used dynamic viewports with 78.1×49.1 degrees of FOV and a resolution of 1920×1080.
It was asked if we consider HMD viewing, and commented that with HMD viewing, the same problem that we had before with people watching different content still exists. So HMD viewing may not be suitable.
It was agreed to use viewport viewing in verification tests.
It was suggested to describe the 360 test procedure in writing, including sequence selection, rate points, viewport selections, and timeline.
Agreed.
A complete inventory of 360° video sequences on our ftp repository (ftp://ftp.ient.rwth-aachen.de) can be found in Annex A. The following candidate list was selected based on their resolutions and discussed in this side activity.
Candidate sequence list for preliminary simulation
Sequence | Frame rate | Resolution | Provider | BD | Camera | CTC | Notes | |
6 | skateboarding | 60 | 4096x2048 | GoPro | 8 | static | no | |
12 | GT_Sheriff | 30 | 4320x2160 | Nokia | 8 | static | no | |
13 | basketball | 30 | 8192x4096 | GoPro | 10 | static | no | |
16 | jam_session | 30 | 8192x4096 | GoPro | 8 | static | no | “holes” in sky/ground |
17 | SkateboardTrick | 60 | 8192x4096 | GoPro | 8 | static | no | |
18 | Train | 60 | 8192x4096 | GoPro | 8 | static | no | “holes” in sky/ground |
28 | BranCastle | 30 | 6144x3072 | GoPro | 8 | moving | no | shaky |
29 | Landing | 30 | 6144x3072 | GoPro | 8 | moving | no | Very fast |
30 | SkateBoardAtBridge | 30 | 6144x3072 | GoPro | 8 | moving | no | |
31 | HarborBiking | 30 | 8192x4096 | InterDigital | 8 | moving | no | |
32 | KiteFliteWalking | 30 | 8192x4096 | InterDigital | 8 | moving | no |
It was commented that having small “holes” in the sky and ground is not an issue if the content is otherwise suitable, because we could design viewports that avoid the “holes.”
During email discussion, it was mentioned that sequences #28 (BranCastle) and #29 (Landing) are shaky and/or contain very fast motion, making them uncomfortable to watch. This was confirmed during the discussion.
It was agreed to not consider sequences #28 and #29 due to undesirable content characteristics.
It was agreed to encode all remaining 9 sequences, numbered 6, 12, 13, 16, 17, 18, 30, 31, 32 in Table 1.
360 video CTC uses the following configurations as “anchor” and “test”:
- codec only comparison: VVC in PERP format vs HEVC in PERP format
- codec + projection comparison: VVC in PHEC format vs HEVC in cubemap (CMP) format
It was agreed to use the same configurations in the preliminary encoding round.
It was asked if also a VR display such as HMD would be used? This could hypothetically be done in a later phase, when a good method of judging the quality of experience of VR would exist. Not in short term, but could be written in the verification test plan as a longer-term goal.
It is also asked if any HDR-based 360 test material would be available. This was not known.