Opened 5 years ago
Closed 5 years ago
#840 closed defect (fixed)
CCALF with Tiles, Slices, Virtual Boundaries
Reported by: | bheng | Owned by: | |
---|---|---|---|
Priority: | minor | Milestone: | VTM-8.0 |
Component: | VTM | Version: | VTM-7.1 |
Keywords: | Cc: | ksuehring, XiangLi, fbossen, jvet@… |
Description
The CCALF code in the "isCrossedByVirtualBoundaries" section of AdaptiveLoopFilter::ALFProcess( ) seems to be incorrect.
First, the pointer to the luma source block is derived as follows, which uses the chroma size and is missing an xPos/yPos offset.
const Area blkSrc(0, 0, w >> chromaScaleX, h >> chromaScaleY);
I believe changing this to something like the following may work:
const Area blkSrc(xPos, yPos, width, height);
Second, the CCALF code seems to ignore all the boundaries. The original luma buffer is used rather than a extracted/padded version. So, the filtering uses luma pixels across these boundaries.
Change history (2)
comment:1 Changed 5 years ago by fbossen
comment:2 Changed 5 years ago by XiangLi
- Milestone set to VTM-8.0
- Resolution set to fixed
- Status changed from new to closed
The call to CCALF does indeed not use the correct parameters in that case. Can you have a look at https://vcgit.hhi.fraunhofer.de/jvet/VVCSoftware_VTM/merge_requests/1280 ? The luma input to CCALF is supposed to be the same as the input to the regular luma ALF. Thanks.