Back to Search Document details
6th Meeting: Hobart, March 2017 2017-04-02 04:12
AHG8: Adjusted cubemap projection for 360-degree video
Abstract
This contribution proposes an adjusted cubemap projection for coding of 360-degree video. The proposed projection targets uniform sphere sampling with limited texture deformation. It uses a simple function to perturb the cubemap lookup vector. The proposed projection can reuse existing cubemap hardware lookup for efficient rendering. The proposed scheme provides -11.3% RA coding gain (E2E WS-PSNR).
JVET-F0025 AHG8: Adjusted cubemap projection for 360-degree video [M. Coban, G. Van der Auwera, M. Karczewicz (Qualcomm)]

This contribution proposes an adjusted cubemap projection for coding of 360-degree video. The proposed projection targets uniform sphere sampling with limited texture deformation. It uses a function to perturb the cubemap lookup vector. The proposed projection can reportedly reuse existing cubemap hardware lookup for efficient rendering. The proposed scheme reportedly provides −11.3% RA coding impact (E2E WS-PSNR).

The rendering is performed in two steps: first mapping the adjusted cubemap into the conventional cubemap, and then performing conventional cubemap rendering.

No additional interpolation is performed, just re-computing the position.

According to several experts’ opinion, the method is not actually compatible with existing GPUs.

It was asked what is the end-to-end PSNR? This was not known, but should be comparable.

Decision (SW): Include in 360Lib software.

It was noted that this projection format was also requested to be included in OMAF; therefore OMAF should exercise an “evil viewport” experiment. The evil viewports of this format would be around the same vertices as in CMP. The effects might be different, since the density of samples is also varied close to the boundaries. Subjective testing was to be performed by the OMAF activity.

JVET-F0025 AHG8: Adjusted cubemap projection for 360-degree video [M. Coban, G. Van der Auwera, M. Karczewicz (Qualcomm)]

Include in 360Lib.

Decisions
noted
It was noted that this projection format was also requested to be included in OMAF; therefore OMAF should exercise an “evil viewport” experiment. The evil viewports of this format would be around the same vertices as in CMP. The effects might be different, since the density of samples is also varied close to the boundaries. Subjective testing was to be performed by the OMAF activity.
Include in 360Lib.
Citation