Back to Search Document details
39th Meeting: Daejeon, KR, March 2025 2025-03-26 09:41
Non-EE2: TIMD-BV extension with enhanced IntraTMP merge list
Abstract
The block vector-based prediction is highly used for intra blocks in ECM. In addition to the regular block-vector based prediction modes of IBC and IntraTMP, the block vectors are used in SGPM, DIMD and OBIC mode to enhance the prediction. This contribution further proposes using block vectors for TIMD mode on top of the TIMD/TIMD-SAD mode. Furthermore, an enhanced IntraTMP merge list sampling strategy is introduced in order to fully explore the potential of the IntraTMP-merge list.
JVET-AL0195 Non-EE2: TIMD-BV extension with enhanced IntraTMP merge list [J. Fu, J. Zhang, Y. Zhao, S. Ma (PKU), C. Huang (ZTE), K. Naser, M. Radosavljević, S. Puri, T. Dumas (InterDigital), D. Ruiz Coll, J.-K Lee (Ofinno)] [late]

The block vector-based prediction is highly used for intra blocks in ECM. In addition to the regular block-vector based prediction modes of IBC and IntraTMP, the block vectors are used in SGPM, DIMD and OBIC mode to enhance the prediction. This contribution further proposes using block vectors for TIMD mode on top of the TIMD/TIMD-SAD mode. Furthermore, an enhanced IntraTMP merge list sampling strategy is introduced in order to fully explore the potential of the IntraTMP-merge list.

Experimental results for the proposed method, implemented on top of the ECM-16.1, are reported as follows:

AI: Overall {-0.08%, -0.06%, -0.01%, 101.1.%, 101.0%}.

Proposal contains two elements: 1) Block vector-TIMD and 2) enhanced IntraTMP merge list, which can be exercised separately.

Some complexity reduction (refinement window is dynamically adjusted based on template cost) is performed in this proposal.

It was commented by several experts that the gain is interesting. To potentially further reduce runtime, the proponent commented that the refinement window adjustment parameter can be further fine-tuned.

The block vector-TIMD aspect of this proposal has some interaction with JVET-AL0174, this is addressed in the combination test in JVET-AL0290.

Cross checker confirmed the reported results (performance and runtime).

See the notes under JVET-AL0290.

Decisions
See the notes under JVET-AL0290.
Citation