| JCTVC-M0386 Cross-check of ‘AHG5: using zero-depth chroma transform unit for 4:2:2 format’ (M0099) by LGE [C. Rosewarne, M. Maeda (Canon)] [late]
Results match, text agrees with the design. The cross-checker reported that the idea is reasonable, but is unsure if it should be used in this form.
JCTVC-M0136 AHG5: 32x32 transform in chroma for 4:2:2 [V. Kolesnikov, C. Rosewarne, M. Maeda (Canon)]
(Discussion chaired by D. Flynn.)
This contribution presents an approach to the residual quad-tree for the 4:2:2 chroma format that allows a 32x32 transform to be used into the chroma channel. At present, the 32x32 transform is used in chroma channels for the 4:4:4 chroma format. It is asserted that reusing the transform logic required for the 4:4:4 chroma format in the 4:2:2 chroma format is beneficial. Simulation results show a gains of 0.0%, 0.3% and −0.3% in luma, chroma for AI Main tier in 4:2:2, 0.0%, 0.4%, 0.4% in luma and chroma for RA Main and 0.0%, 0.7%, 0.9% in luma and chroma for LD Main.
For a 64x64 CU, there is an implied initial split in the residual quadtree due to the maximum transform size of 32x32. In the 4:2:2 case, the derivation of the chroma RQT from luma causes a TU size of 16x16 to be used. This contribution removes the implicit inference of the transform split flag in this case, requiring it to be signalled explicitly for chroma. The rationale being that if a decoder implementation uses 4:4:4, then it will have the 32x32 transform anyway, so this may as well be made available in 4:2:2. However, this assumes that there won't be a 4:2:2 specific profile as in AVC.
It was remarked that, although 32x32 chroma transforms are not currently present in the 4:2:2 design, if they were, L0099 could achieve the same effect more generally.
Before deciding the issue of whether this is the way to achieve such an outcome, it should be ascertained if a 32x32 chroma transform is desirable in 4:2:2.
Decision: No 32x32 chroma transform for 4:2:2 (through lack of input).
No action.
Dostları ilə paylaş: |
