JVET-Q0227 Dependent Quantization with QP offset adaptation [P. de Lagrange, F. Hiron, F. Le Léannec, E. François (InterDigital)]
This contribution relates to Dependent Quantization (DepQ). When DepQ applies, an qP offset of +1 is added to the qP value when applying the scaling process for transform coefficients. No offset is used when DepQ is off. In this contribution, it is proposed to adapt the offset value depending on the qP value. The reported psnr-Y, U, V BD-rate variations compared to VTM7.0 are -0.07%, 0.29%, 0.28% for the AI configuration, -0.07%, -0.23%, -0.34%for the RA configuration, -0.18%, 0.08%, -0.17% for the LDB configuration.
The offset is set to +2 for QP>30, +1 for 30 >= QP > 23, +0 otherwise.
It is pointed out that some of the effect could be due to the change in lambda, and that basically also an encoder could just use QP control to achieve this effect. The best choice could also be sequence dependent.
The gain is relatively small.
More investigation beyond CTC might be necessary to better study the effect.
No action.