Back to Search Document details
13th Meeting: Marrakech, January 2019 2019-01-04 15:36
CE2: Affine motion predictor pruning (test 2.2.8)
Authors: Antoine ROBERT Technicolor F. Leleannec Technicolor Tangi Poirier Technicolor Franck Galpin Technicolor
Abstract
This contribution describes a simple motion predictor pruning process to be applied in Affine AMVP and Affine Merge modes to avoid redundancy in the lists of predictors and to be close to the AMVP processes. The proposed process has been implemented in the JVET VTM3.0. The reported simulation results show an average luma BD-rate gain of -0.02% in RA configuration and the same results in LDB configuration. The encoding and decoding times stay identical to the VTM-3.0 ones. The worth case complexity in terms of integer comparisons required by the pruning is increased to 18 comparisons per reference frame in Affine AMVP and by 14 comparisons in addition to the 78 actual ones in Affine Merge mode.Combined with the proposal of JVET-M0281, corresponding to CE4.1.5, that performs motion predictor pruning for AMVP and Merge modes, the reported simulation results show an average luma BD-rate gain of -0.04% in RA configuration and of -0.03% in LDB configuration. The encoding and decoding times here also stay identical to the VTM-3.0 ones. And, the worth case complexity in terms of integer comparisons required by the pruning is further increased by 24 comparisons in addition to the 174 actual ones in Merge mode.
Meeting Notes
JVET-M0282 CE2: Affine motion predictor pruning (test 2.2.8) [A. Robert, F. Le Léannec, T. Poirier, F. Galpin (Technicolor)]

This proposal includes two aspects. First, before any pruning operation takes place the involved motion vectors are computed in their final representation. This means the MVs are set to the appropriate precision level, rounded to the right motion vector resolution or clipped, before being compared to candidates already present in the concerned list. Second aspect consists in adding some pruning operations between candidates to improve the trade-off between coding efficiency and complexity.

References:
JVET-L1010 JVET-L1022 JVET-M0281
Cited:
JVET-M0022 JVET-M0283
PATENTS:
WO2026026926A1 0.34 2026-02-05 CN113287317A 0.32 2021-08-20 WO2025002293A1 0.30 2025-01-02 US20230353737A1 0.28 2023-11-02 WO2024104476A1 0.26 2024-05-23 CN112913247A 0.24 2021-06-04 JP2024008948A 0.22 2024-01-19 JP7648722B2 0.20 2025-03-18
ARCHIVE v2 :
CE2.2.8.a_tester.xlsm CE2.2.8.b_tester.xlsm CE2.2.8.c_tester.xlsm CE2.2.8.d_tester.xlsm JVET-M0282_v2.docx JVET-M0282_v2.pptx
Decisions
This proposal includes two aspects. First, before any pruning operation takes place the involved motion vectors are computed in their final representation. This means the MVs are set to the appropriate precision level, rounded to the right motion vector resolution or clipped, before being compared to candidates already present in the concerned list. Second aspect consists in adding some pruning operations between candidates to improve the trade-off between coding efficiency and complexity.
Citation