JVET-L0362 Quantization parameter signalling [Y. Zhao, H. Yang, J. Chen (Huawei)]
The concept of quantization group (QG) was introduced in HEVC for signalling QP for non-overlapped square regions in a coding picture. The QG size can be decided by an encoder for the trade-off between QP signalling overhead and picture quality improvement. A quantization group may contain multiple CUs while CU larger than quantization group is also supported. However, the concept of quantization group breaks in current design of VVC in that a quantization group may contain multiple incomplete CUs, due to non-square CUs. Two mechanisms are introduced in this contribution to extend the concept of quantization group in the context of QT-MTT partitioning framework. The two methods are tested with CU-level perceptually-optimized delta QP calculation as described in CE7.2.3, with MaxCuDQPDepth = 2 and 4.
Q: why use depth instead of area? Several experts commented that it may be better to use area instead of depth. It was commented that there are some complications with TT splits when considering area.
Two methods are proposed. They differ in how they define depth to QP signalling purpose. The first method considers QT, BT and TT splits to determine depth. The second method considers only QT splits.
It was remarked that method 1 is closer to what is already implemented in VTM.
Decision: Adopt method 1 (depth is QT depth + BT depth)