Back to Search Document details
16th Meeting: Geneva, October 2019 2019-10-08 09:02
CE5-related: Simplified CCALF
Authors: Olena Chubach MediaTek C.-Y. Lai MediaTek C.-Y. Chen MediaTek T.-D. Chuang MediaTek Y.-W. Huang MediaTek Shawmin Lei MediaTek
Abstract
In CE5-2, the filtered result of cross-component adaptive loop filter (CCALF) is added on top of the chroma adaptive loop filter (ALF) output. To support CCALF, the number of multiplications per pixel is increased from 15 to 22 (46.7% increase compared to VTM6.0 ALF). In this contribution, two simplifications are proposed in order to reduce the computation complexity of CCALF tested in CE5-2.
Meeting Notes
JVET-P0165 CE5-related: Simplified CCALF [O. Chubach, C.-Y. Lai, C.-Y. Chen, T.-D. Chuang, Y.-W. Huang, S.-M. Lei (MediaTek)]

In CE5-2, the filtered result of cross-component adaptive loop filter (CCALF) is added on top of the chroma adaptive loop filter (ALF) output. To support CCALF, the number of multiplications per pixel is increased from 15 to 22 (46.7% increase compared to VTM6.0 ALF). In this contribution, two simplifications are proposed in order to reduce the computation complexity of CCALF tested in CE5-2.

  1. It is proposed to select CCALF, chroma ALF, or ALF-off for each chroma coding tree block (CTB), and signal the selected option explicitly. As a result, the number of multiplications per pixel is reduced from 22 to 19 (26.7% increase compared to VTM6.0 ALF).
  2. It is proposed to update the filter shape of the CCALF and to align to that of chroma ALF. The second simplification contains four subtests: a) all CCALF coefficients can have different values; b) top and bottom corresponding coefficients are equal, and left and right coefficients are equal; c) top, bottom, left, and right corresponding coefficients are all equal.

Simulation results with chroma gains shifted to luma are summarized as follows.

The YCbCr BD-rate (Y-BD-rate*0.8 + Cb-BD-rate*0.1 + Cb-BD-rate*0.1, according to AHG13) of the CE5-2.1 CCALF is -0.85% for RA while the number of multiplications in ALF is increased by 46.7% (comparing “VTM6.0 ALF + CCALF” to “VTM6.0 ALF”).

Method 1 preserves 78.3% of CE5-2.1 CCALF’s coding gain while saving 43% of its multiplications.

Method 2a preserves 99.9% of CE5-2.1 CCALF’s coding gain while saving 7% of its multiplications.

Method 2b preserves 95.5% of CE5-2.1 CCALF’s coding gain while saving 21% of its multiplications.

Method 2c preserves 89.7% of CE5-2.1 CCALF’s coding gain while saving 29% of its multiplications.

Method 1+2a preserves 80.7% of CE5-2.1 CCALF’s coding gain while saving 50% of its multiplications.

Method 1+2b preserves 75.4% of CE5-2.1 CCALF’s coding gain while saving 64% of its multiplications.

Method 1+2c preserves 71.2% of CE5-2.1 CCALF’s coding gain while saving 71% of its multiplications.

Proponents suggested there was no need to review this.

References:
JVET-N1010 JVET-O0636
Cited:
JVET-P0016 JVET-P0739 JVET-P0923 X JVET-P0949 X JVET-P1008
PATENTS:
20250220169 0.92 WO/2024/012168 0.92 WO/2025/001782 0.90 WO/2026/016800 0.90 20230156234 0.90 20230171403 0.88 WO/2025/152997 0.70 WO/2025/152690 0.33
ARCHIVE v3 :
JVET-P0165-v1.docx JVET-P0165-v2-clean.docx JVET-P0165-v2.docx JVET-P0165-v3-clean.docx JVET-P0165-v3.docx JVET-P0165_Presentation.pdf JVET-P0165_Spec_Method1+2a.docx JVET-P0165_Table1.xlsm JVET-P0165_Table2.xlsm JVET-P0165_Table3.xls JVET-P0165_Table4.xls JVET-P0165_Table5_partial.xls JVET-P0165_Table6_partial.xls JVET-P0165_Table7_partial.xls JVET-P0165_Table8_partial.xls
Decisions
Proponents suggested there was no need to review this.
Citation