JVET-AC0198 Non-EE2: IntraTMP with multiple modes [P.-H. Lin, J.-L- Lin, V. Seregin, M. Karczewicz (Qualcomm)]

This contribution proposes to introduce additional intra template matching (IntraTMP) modes to enhance the coding performance without increasing much complexity for both encoder and decoder. The experimental results show that the luma BDrate gain is -0.13% with 101%/101% encoding/decoding time in AI condition on top of ECM-7.0.

Three modes: left template, above template, and L-shaped

Is the encoder run time reliable? There is reduction in some classes, and increase (up to 8%) in particular for the classes A.

It was pointed out that there is a division in the determination of the fusion weights. It was answered that similar determination methods are already in other places as of now.

No change to chroma

What is the individual benefit of the two aspects (adaptation of templates, modification of the fusion)? Should also be studied separately.

Cost for the three template modes is partially e-used, as the left and above templates are subsets of the L-shape.

The modified fusion is only used for L-shape.

It was agreed to study this in an EE.