Back to Search
Document details
CE2-related: MV Derivation for Affine Chroma
Abstract
In this contribution, it is proposed to simplify the MV derivation process for chroma sub-block of an affine CU. The size of chroma sub-block is 4x4 samples, which corresponds to 8x8 luma samples. To obtain the MV for the chroma sub-block, the MVs of each of the four 4x4 luma sub-blocks are averaged and rounded. Furthermore, the rounding process employed for computing the average of the four MVs is not symmetric for positive and negative numbers, because the current implementation uses right shifting by 2 after adding an offset of 2 irrespective of sign of the input. This contribution proposes several options simplify the MV derivation process and address the asymmetric rounding issue: 1) Use only the top-left luma sub-block’s MV as the representative MV 2) Use the average of the top-left and bottom-right luma sub-blocks’ MVs as the representative MV 3) Modify the current averaging process of four luma sub-block MVs to a symmetric one 4) Use the average of the top-left and bottom-right luma sub-blocks’ MVs as the representative MV while using symmetric rounding. All the proposed methods show marginal performance change.
JVET-M0192 CE2-related: MV Derivation for Affine Chroma [A. Tamse, M. W. Park, K. Choi (Samsung)]
References:
PATENTS:
US20220248028A1
1.00
2022-08-04
US12167020B2
0.89
2024-12-10
BR122025014629A2
0.87
2025-11-25
EP3972250A1
0.87
2022-03-23
US20250047846A1
0.34
2025-02-06
US11943469B2
0.30
2024-03-26
US20220078458A1
0.28
2022-03-10
US20220239900A1
0.24
2022-07-28
Decisions
JVET-M0192 CE2-related: MV Derivation for Affine Chroma [A. Tamse, M. W. Park, K. Choi (Samsung)]
Citation