JVET-Q0474 AHG15: defining QP at TU level [P. de Lagrange, F. Le Léannec, F. Urban, K. Naser (InterDigital)]
This contribution proposes to define QP at the same level QP-delta is signalled, i.e. TU level instead of CU level. This changes the way QP is assigned to leading TUs that do not carry coded coefficients within a CU that carries a QP-delta, removing QP-delta back-propagation and aligning the process with leading CUs without coded coefficients within a QG. This is said to be a cleanup, which is mostly a semantic change, and removes the need for fixes adopted from JVET-O0046 and JVET-P0436 by solving the root cause.
The method would change the current approach, where always the same QP is used over the whole CU, by associating the QP with TU
It would keep the following concepts unchanged
- QG can not be smaller than a CU
- QP can only change once in a QG
As a consequence, Luma TBs within ISP could have different QPs
Question is raised if QPs are still stored with 4x4 for deblocking? Proponent says yes.
Not clear what the advantage is? Proponent claims that the current design requires back propagation of QP for the deblocking of TBs that belong to the same CU but have no transform coefficients. This problem was however solved in last meeting by sending delta QP up front.
Not obvious that there is a problem to be solved, and the proposal would require a substantial amount of semantics changes which would change the normative behaviour.
No interest for adoption.