JVET-Q0322 CE1-related: Long luma deblocking filter decision modification [H.-B. Teo, C. Wang, C.-S. Lim (Panasonic)]
It is claimed that sometimes filtering for long luma deblocking is applied, although it should not be. The wrong decision is asserted to sometimes cause visual artefacts. This contribution proposed to fix the problem by changing the position of the samples used for the gradient checks when max filter length equal to 7.
The BDR impact versus VTM-7.0 is 0.00%, 0.01%, -0.02%, 0.03% for AI/RA/LDB/LDP.
It was commented that the solution adopted in CE1-1 targets solving the same problem of oversmoothing of the long filter. It is further pointed out that the suggested approach does not consider the gradient between p0/p1 and q0/q1 which are rather important in deblocking (gradient close to the block edge) which might cause other problems.
It would be too much of an effort, considering the short time until finalization of the standard, to investigate this in detail.
No action was taken on this.