JVET-H0048 AhG8: Dynamic viewport based subjective evaluation of Balboa and Landing [P. Hanhart, Y. He, Y. Ye (InterDigital)]

This document reports results of a subjective evaluation comparing dynamic viewports for the newly adopted Balboa and Landing sequences. The HM and JEM CfP anchors were used in the evaluation, as well as the updated Landing sequence (JVET-H0021), which was encoded with HM-16.16 and JEM-7.0 in the padded equirectangular projection (PERP) format under CfP coding conditions for the current Landing sequence. The double stimulus impairment scale (DSIS) method was used to assess the subjective quality of these sequences coded using HM and JEM. Results show that JEM achieves some visual quality improvements over HM across the selected target bit rates. Finally, some comments regarding the Landing sequence are made based on the feedback from the viewers.

Experts viewing was performed (9 participants), using the 5 grade MOS scale.

For Balboa, the range of rates reportedly seems OK, differentiation between JEM and HM with non-overlapping confidence intervals was reported up to the highest bit rate.

Landing1 is rated at highest rate with a MOS of 3.x. Viewport 1 seems to be inappropriate, as high camera motion in the end may be disturbing. It was pointed out in the discussion that for 360° video with a high amount of camera motion that too much additional variation in dynamic viewport should be avoided.

Landing 2 is reportedly somewhat better. It was however observed that in the case of both JEM and HM coding the variation over rates 1–4 is not large for viewport 2 (looking up to sky). Nevertheless, differences between HM and JEM are still visible.

If we make a reduction to 5 sequences, Landing is probably a candidate for being dropped.

It was later decided that Landing will not be included in the CfP.