JVET-F0083 AHG8: Subjective test pilot study of 360º video projection/packing formats [Z. Deng, L. Xu, J. Boyce (Intel)] [late]

A mini subjective test pilot study was conducted for the two methods described in F0021.

1. Static viewport with forced discontinuous edges at picture boundary
2. Static viewport with forced vertexes

The first method applies pre-rotation to force discontinuous edges at the picture boundary to be placed in a static viewport. The second proposed method similarly requires pre-rotation to force a static viewport center to a vertex inside the picture.

By pre-rotating each sequence prior to projection/packing format mapping and encoding, discontinuous edges will be included in the static viewport. After lossy encoding, especially with high QP, the compression distortion would make the discontinuous edges visible in the decoded static viewport. In the reported mini subjective viewing test pilot study, these seam/edge artefacts were observed in the decoded viewport for both Method 1 and Method 2, and the artefacts were observed with different extent across different projection formats. To evaluate the visual quality associated with various projection/packing formats, it was suggested to be meaningful to include the proposed pre-rotation based methods into the common test conditions and evaluation procedures for 360º video coding and that the proposed methods could also be used for the quality assessment of those techniques that would possibly be developed to alleviate or eliminate those seam artefacts.

Method 1 defines position at picture boundary, method 2 within the picture.

Subjective testing was performed with three subjects, and a MOS scale was obtained. They detected visible artefacts for several projection formats, but the correlation with viewport PSNR are rather low.

During the discussion, it had also been suggested that testing various viewports and measuring PSNR might be an alternative, but results show that viewport PSNR does not reliably detect the seam problems.

Both methods seemed needed, since some projection formats might have more critical areas at boundary, others in the inner part.

From video demonstrations given during the presentation, obvious visibility of artefacts was confirmed. It seemed, however, unclear whether this could be avoided by some encoder optimization (e.g. decreasing QP at face boundaries).

See further notes under JVET-F0021.