JVET-Y0076 Non-EE2: Template Matching-based OBMC Design [Z. Lv, C. Zhou, J. Zhang (vivo)]

This contribution presents a method of template matching-based OBMC design, where the prediction value of CU boundary samples derivation approach is decided according to the template matching costs. Instead of directly using the weighted prediction, the proposed method chooses from three approaches for each block, including using current block’s motion information only, or using neighbouring block’s motion information as well with one of the two blending modes.

On top of ECM-3.1, the simulation results of the proposed method are reported as below:

Test 1. RA: -0.13%, -0.18%, -0.10%, 114%, 118%.

Test 2. RA: -0.13%, -0.21%, -0.16%, 100%, 102%.

The RA results above were not obtained by correct CTC, but new incomplete results with CTC reportedly showed practically the same performance.

The template matching used here is somewhat different from other approaches in the ECM, may have some commonality with the template used in AMVR

It was asked if a similar effect (establishing rules whether to perform OBMC or not, method of blending, etc.) really needs template matching, or could be determined by other criteria.

This is generally asserted as having interesting aspects.

It was agreed to investigate Test 2 in an EE.