JVET-P0552 AHG12/Non-CE5: Extending slice boundary processing for adaptive loop filter for raster scanned slices [N. Hu, V. Seregin, M. Coban, A. K. Ramasubramonian, M. Karczewicz (Qualcomm)]
Contributions JVET-P0121, JVET-P0156, JVET-P0492, JVET-P0552 are related.
To handle unsupported cases in VVC draft 6, it is proposed to convert such cases to a case, which is already supported, such as to treat the above CTU boundary or the left CTU boundary as a slice boundary even some samples belong to the same slice.
The problem is converted into the case when the top CTU is in a different slice such that mirrored padding is applied along the top CTU boundary.
It was agreed to add text to the specification as a fix as given below:
- The variables ac[ x ][ y ], sumH[ x ][ y ], sumV[ x ][ y ], sumD0[ x ][ y ], sumD1[ x ][ y ] and sumOfHV[ x ][ y ] with x, y = 0..( CtbSizeY − 1 ) >> 2 are derived as follows:
The variables clipLeftPos, clipRightPos, clipTopPos and clipBottomPos are derived by invoking the ALF boundary position derivation process as specified in clause 8.8.5.5 with ( xCtb, yCtb ), ( x << 2, y << 2 ).
Proposals JVET-P0156 and JVET-P0552 uses top padding for the top-left block when top-left sample(s) are unavailable. JVET-P0492 uses a repetitive padding from the top-left corner sample.
The definition of a sample being unavailable uses a sum of the top and left offset in JVET-P0552 and JVET-P0492, and two separate checks (left and right) in JVET-P0156. The condition in JVET-P0552 is tighter.
Since the repetitive padding was recommended (see discussion on padding), proposals JVET-P0156, JVET-P0552 would use the repetitive padding instead.
BoG recommendation: It was agreed to use the check for the unavailable samples as in JVET-P0552 (and same as in JVET-P0492).
It was agreed that the repetitive padding is used for all the top samples, when there are unavailable samples. This is similar to what was proposed in JVET-P0156 and JVET-P0552 except that the mirrored padding is replaced by the repetitive padding according to the discussion earlier. This means a top horizontal line in the CTU is used to pad top samples. This applies to the top-left unavailable case, the bottom-right case would use symmetrical operations.
Further discussed on Thursday, Oct. 10, at 1820.
It was mentioned that current VTM would use sample (0,0) to fill unavailable samples to the top-left (as in JVET-P0492) for the top-left corner of the proposal.
It was mentioned that even the current design may not be good.
It was proposed by Minhua (Broadcom) to use horizontal repetitive padding for filling unavailable samples. No text or software are available though.
Support has been expressed for the proposal from Broadcom.
Support has also been expressed for using the top left sample to fill the unavailable samples JVET-P0492. It was commented that JVET-P0492 would create a special case.
Recommended: Adopt padding as in Broadcom verbal proposal that uses horizontal repetitive padding. Bytedance volunteered to provide specification text and the software.
Clarification: The unavailable samples are determined as in JVET-P0552 / JVET-P0492.