JVET-R0184 AHG9/AHG12: On reference picture wraparound for subpictures [S. Paluri, Hendry, S. Kim (LGE)]
Discussed in AHG Session 1.15 on Monday 13 April (chaired by GJS & YKW).
In the current VVC draft, reference picture wraparound for subpictures is only possible when the subpic_treated_as_pic_flag is equal to 0. This contribution asserted that the reference picture wraparound functionality could be extended for cases when subpic_treated_as_pic_flag equal to 1 with the subpicture width is equal to the picture width without having to have additional signalling. This contribution enables this functionality semantically.
In this contribution, the following is proposed:
- Allow reference picture wraparound for subpicture with subpic_treated_as_pic_flag[ i ] equal to 1 when the width of the subpicture is the same as the width of the picture, without additional signalling.
- At slice level, derive the boundaries to be applied for decoding of blocks in the slice. If the slice belongs to an independently coded subpiture, use the subpicture boundary; otherwise, use picture boundary. It is remarked that this 2nd proposal item can be considered as editorial updates.
Examples shown in JVET-Q0403 were discussed. Two examples are shown below.
Case 1
Wrap off | Wrap off |
Wrap on | |
Wrap on | |
The above example would have different behaviour in different subpictures. The next figure would not.
Case 2
Wrap on |
Wrap on |
The second case could be used for, e.g., top-bottom stereoscopic ERP coding (or segmenting a single ERP).
Both of these are currently disallowed with subpictures treated as picture boundaries.
It was commented that we should avoid potential confusion in what is allowed and how it operates.
It was suggested to have wrap-around be a whole-picture property. This would invoke wrap-around as follows, with wrapping at the picture boundaries (thick boundaries, independent of the subpicture layout), as in case 3.
Case 3
The group seemed inclined to consider only the second of these three variations (case 2). The case just above and the case below would be prohibited:
Case 4
Wrap on |
Wrap off |
For signalling, wrap is in the SPS and the wrap offset is signalled in the PPS.
AHG Recommendation (cleanup): Allow case 2 (only). Text was later provided in a -v2 revision and confirmed to be OK on Wednesday.
Discussion for AHG Session 1.15 ended on Monday 13 April at 1525 UTC.