Text/SW mismatch CIIP deblocking (8.3.3.5)

[fbossen]

In the presence of CIIP-coded blocks, boundary strength setting is governed by:
– Otherwise, if the block edge is also a coding block edge and the sample p0 or q0 is in a coding block with ciip_flag equal to 1, bS[ xDi ][ yDj ] is set equal to 2.

This appears to be inconsistent with the SW, which I believe does not have the restriction on the block edge being a coding block edge. For example, in the case where the max transform size is 32 and a CU is larger than 32 in either dimension, there are transform block edges internal to the CU, and as far as I can tell, the SW assigns a value 2 to bS for these internal edges if the CU uses CIIP.

To align with the software, the text should be modified as:
– Otherwise, if the sample p0 or q0 is in a coding block with ciip_flag equal to 1, bS[ xDi ][ yDj ] is set equal to 2.

[jlchen]

It sounds reasonable to me matching the specification text to software for this issue. Can other deblocking experts confirm this change?

[fbossen]

VTM9 bitstream with CIIP and max transform size set to 32

[fbossen]

Since none of the VTM9 conformance streams seem to exercise this case, a bitstream that does is provided.

[Kenneth]

Good catch Frank. Make sense to align text with the software as you suggest.

[bbross]

Thanks! This will be fixed in JVET-S2001-vE.