JVET-M0091 AHG10: Clean-up and finalization of perceptually optimized QP adaptation method in VTM [C. Helmrich (HHI)]

Presented Thursday 17 January 2020 (chaired by F. Bossen).

This contribution proposes a clean-up and a completion of the perceptually optimized QP adaptation (QPA) algorithm already integrated into the VTM codec software. Specifically, the following points are addressed:

1. for HD and smaller input sequences, the previously employed reduction of the CTU size is removed
2. for HD and smaller input, a depth-1 QPA (4 QPs per CTU) is used instead of the CTU size reduction
3. the QPA parameter _pic is now defined as a function of the picture size instead of by a case-statement
4. an extension of the QPA algorithm for better visual handling of CTUs with glaring colors is provided
5. some remaining obsolete QPA related is code removed and some DC offset calculations are unified.

The proposed changes, whose integration into the next VTM software version is suggested, result in slightly reduced bitstream sizes for HD or smaller sequences and the Campfire UHD sequence and reportedly provides between 1.7 and 2.5% additional luma BD-rate gain on the random-access coded sequences of the SDR common test conditions (CTC). However, the proposal does not affect the CTC since QPA is off by default.

Decision (SW): Adopt.