Back to Search Document details
14th Meeting: Geneva, March 2019 2019-03-20 09:01
CE3: Simplification of LM mode (CE3-1.6)
Authors: Yi-Wen Chen Kwai Hong-Jheng Jhu Kwai X. Wang Kwai
Abstract
Reported in this document is the results of CE3-1.6 on the simplification of LM modes. There are six sub-tests under this CE test. In the CE3-1.6.1, a modification to the cross-component linear model (CCLM) deriving method is presented. To determine the α and β for the CCLM model, currently in VVC the minimum and maximum luma samples are searched among the down-sampled top and left neighbouring reconstructed luma samples. α and β are then determined by the linear model formed from the minimum/maximum down-sampled luma samples and the corresponding chroma samples. In this proposal, it is proposed to search the minimum and maximum luma samples among the top and left neighbouring reconstructed luma samples without the down-sampling process. After the minimum and maximum luma samples are identified, their corresponding down-sampled luma samples are then generated to derive the α and β. Simulation results reportedly show BD-rates on Y, Cb and Cr components are 0.02%, 0.19%, and 0.22% for AI configuration over VTM-4.0, with 100% encoding time and 100% decoding time. The combination of the proposed simplification and the other CE tests which reduce the number of samples used to derive the LM model are also tested in CE3-1.6.2 to CE3-1.6.6. The simulation results reportedly show that the proposed simplification has the similar behavior in the combination tests in terms of coding efficiency.
Meeting Notes
JVET-N0321 CE3: Simplification of LM mode (CE3-1.6) [Y.-W. Chen, H.-J. Jhu, X. Wang (Kwai Inc.)]

Reported in this document is the results of CE3-1.6 on the simplification of LM modes. There are six sub-tests under this CE test. In the CE3-1.6.1, a modification to the cross-component linear model (CCLM) deriving method is presented. To determine the α and β for the CCLM model, currently in VVC the minimum and maximum luma samples are searched among the down-sampled top and left neighbouring reconstructed luma samples. α and β are then determined by the linear model formed from the minimum/maximum down-sampled luma samples and the corresponding chroma samples. In this proposal, it is proposed to search the minimum and maximum luma samples among the top and left neighbouring reconstructed luma samples without the down-sampling process. After the minimum and maximum luma samples are identified, their corresponding down-sampled luma samples are then generated to derive the α and β. Simulation results reportedly show BD-rates on Y, Cb and Cr components are 0.02%, 0.19%, and 0.22% for AI configuration over VTM-4.0, with 100% encoding time and 100% decoding time. The combination of the proposed simplification and the other CE tests which reduce the number of samples used to derive the LM model are also tested in CE3-1.6.2 to CE3-1.6.6. The simulation results reportedly show that the proposed simplification has the similar behavior in the combination tests in terms of coding efficiency.

References:
JVET-M0229 JVET-M1010
Cited:
JVET-N0023 JVET-N0577 X
PATENTS:
US20220017661A1 0.34 2022-01-20 SG191441A1 0.32 2013-07-31 SG191440A1 0.30 2013-07-31 SA91120187B1 0.28 2004-05-09 JPH02500514A 0.26 1990-02-22 BE365077A 0.24 CA1059770A 0.22 1979-08-07 EP1092701A1 0.20 2001-04-18
ARCHIVE v3 :
CE3-1.6.1_tester.xlsm CE3-1.6.2_tester.xlsm CE3-1.6.3_tester.xlsm CE3-1.6.4_tester.xlsm CE3-1.6.5_tester.xlsm CE3-1.6.6_tester.xlsm JVET-N0321.docx JVET-N0321_CombineWithCE3-1.7.xlsm JVET-N0321_CombineWithCE3-1.8.2.xlsm JVET-N0321_r1.docx JVET-N0321_r2.docx JVET-N0321_WD.docx
Decisions
Reported in this document is the results of CE3-1.6 on the simplification of LM modes. There are six sub-tests under this CE test. In the CE3-1.6.1, a modification to the cross-component linear model (CCLM) deriving method is presented. To determine the α and β for the CCLM model, currently in VVC the minimum and maximum luma samples are searched among the down-sampled top and left neighbouring reconstructed luma samples. α and β are then determined by the linear model formed from the minimum/maximum down-sampled luma samples and the corresponding chroma samples. In this proposal, it is proposed to search the minimum and maximum luma samples among the top and left neighbouring reconstructed luma samples without the down-sampling process. After the minimum and maximum luma samples are identified, their corresponding down-sampled luma samples are then generated to derive the α and β. Simulation results reportedly show BD-rates on Y, Cb and Cr components are 0.02%, 0.19%, and 0.22% for AI configuration over VTM-4.0, with 100% encoding time and 100% decoding time. The combination of the proposed simplification and the other CE tests which reduce the number of samples used to derive the LM model are also tested in CE3-1.6.2 to CE3-1.6.6. The simulation results reportedly show that the proposed simplification has the similar behavior in the combination tests in terms of coding efficiency.
Citation