Back to Search Document details
29th Meeting: by teleconference, DE, January 2023 2023-01-16 16:04
AHG12: Using block vector derived from IntraTMP as an IBC candidate for the current block
Authors: J.-K. Lee Ofinno D. Ruiz Coll Ofinno V. Warudkar Ofinno DAMIAN RUIZ COLL
Abstract
This contribution proposes to use a block vector (BV) derived from intra template matching prediction (IntraTMP) for the current block as a block vector predictor (BVP) for the AMVP IBC candidate list. After IntraTMP achieves the best reference block, and the TMP-BV is stored in the AMVP IBC list as a new BVP candidate for the current block. Simulation results based on ECM-7.0 are reported below:
Meeting Notes
JVET-AC0167 AHG12: Using block vector derived from IntraTMP as an IBC candidate for the current block [J.-K. Lee, D. Ruiz Coll, V. Warudkar (Ofinno)]

This contribution proposes to use a block vector (BV) derived from intra template matching prediction (IntraTMP) for the current block as a block vector predictor (BVP) for the AMVP IBC candidate list. After IntraTMP achieves the best reference block, and the TMP-BV is stored in the AMVP IBC list as a new BVP candidate for the current block. Simulation results based on ECM-7.0 are reported below:

AI: Class F: -0.07%, 0.10%, -0.12%, 100%, 97%; Class TGM: -0.04%, -0.00%, -0.02%, 98%, 97%;

RA: Class F: -0.10%, -0.04%, -0.13%, 104%, 102%; Class TGM: 0.03%, -0.10%, 0.01%, 103%, 103%

The proposal replaces the second AMVP BVP predictor with the proposed IntraTMP based BVP. When this candidate is selected, the decoder would need to perform IntraTMP to derive this predictor.

Simulation results are confirmed by cross checker.

It was commented that the gains for the SCC classes are rather low, some loss in class TGM in RA configuration.

It was commented that JVET-AB0061 which was adopted at the last meeting could be related. That contribution uses IntraTMP block vector from the neighbouring blocks, whereas this contribution uses IntraTMP block vector from the current block. JVET-AB0061 achieves 0.04%/0.11% for class F/TGM in AI configuration.

JVET-AB0061 had low gains but was considered simple extension as it does not introduce any additional operations. Further study encouraged, especially in the context of whether the proposed method is beneficial for camera captured content as there is an EE2 test on IBC for camera captured content. Could be candidate for EE at next meeting if benefits are convincing.

Cited:
JVET-AC0283 X
ARCHIVE :
JVET-AC0167_v3.docx JVET-AC0167_v3.pptx JVET-AC0167_v3.xlsm JVET-AC0167_v3_clean.docx ~$JVET-AC0167_v1.xlsm ~$JVET-AC0167_v2.pptx ~$JVET-AC0167_v2.xlsm
Decisions
JVET-AB0061 had low gains but was considered simple extension as it does not introduce any additional operations. Further study encouraged, especially in the context of whether the proposed method is beneficial for camera captured content as there is an EE2 test on IBC for camera captured content. Could be candidate for EE at next meeting if benefits are convincing.
Citation