Back to Search Document details
39th Meeting: Daejeon, KR, June 2025 2025-06-29 05:12
Non-EE2: Adaptive subsampling filter selection for CCLM/CCCM
Authors: yoshitaka kidani KDDI H. Kato KDDI K. Kawamura KDDI
Abstract
This contribution proposes an adaptive subsampling filter selection for the cross-component linear model (CCLM) and convolutional cross-component model (CCCM). Considering that many 4:2:0 color format videos are generated from 4:4:4 color format videos using different subsampling filters, it may not be optimal to utilize only one subsampling filter for CCLM and CCCM in the current ECM. This is because it occurs phase shifts between luma and chroma samples. To align the subsampling filters between the generation process of 4:2:0 color format videos and CCLM and CCCM processes, an adaptive subsampling filter selection method for CCLM and CCCM is proposed. Specifically, in the proposed method, different subsampling filters are adaptively used for the luma subsampling processes of CCLM and CCCM (including intra-CCCM and ALF-CCCM) by reusing the existing SPS flags in ECM, which specify the video color format and spatial allocation of luma and chroma samples. Similarly, the adaptive filter selection method is also introduced into the filter derivation and application processes of inter-CCCM.
Meeting Notes
JVET-AM0196 Non-EE2: Adaptive subsampling filter selection for CCLM/CCCM [Y. Kidani, H. Kato, K. Kawamura (KDDI)]

This contribution proposes an adaptive subsampling filter selection for the cross-component linear model (CCLM) and convolutional cross-component model (CCCM). Considering that many 4:2:0 colour format videos are generated from 4:4:4 colour format videos using different subsampling filters, it may not be optimal to utilize only one subsampling filter for CCLM and CCCM in the current ECM. This is because it occurs phase shifts between luma and chroma samples. To align the subsampling filters between the generation process of 4:2:0 colour format videos and CCLM and CCCM processes, an adaptive subsampling filter selection method for CCLM and CCCM is proposed. Specifically, in the proposed method, different subsampling filters are adaptively used for the luma subsampling processes of CCLM and CCCM (including intra-CCCM and ALF-CCCM) by reusing the existing SPS flags in ECM, which specify the video colour format and spatial allocation of luma and chroma samples. Similarly, the adaptive filter selection method is also introduced into the filter derivation and application processes of inter-CCCM.

On top of the ECM-17.0, the proposed method provides some coding gains without increasing encoder and decoder runtime in specific test sequences of CTC and the draft of Joint Call for Evidence on video compression with capability beyond VVC. These simulation results {Y, U, V, EncT, DecT} for the all-intra configuration are reported below:

  • CTC:
    • CatRobot {-0.07%, -0.99%, -0.97%, 100.5%, 100.0%},
    • FourPeople {-0.01%, -0.12%, -0.40%, 100.5%, 99.7%},
    • SlideShow {-0.24%, -4.09%, -5.01% , 99.6%, 100.1%},
  • Draft of Joint Call for Evidence on video compression with capability beyond VVC:
    • SDR_RA_UHD:
  • CrowdRun {-0.03%, -0.83%, -0.66%, 100.0%, 99.4%},
  • FireDance {-0.14%, -0.29%, -0.12%, 99.6%, 99.0%},
    • SDR_LD_HD:
  • GregoryCactus2 {-0.02%, -0.04%, -0.03%, 100.3%, 99.3%},
    • HDR_RA_8Kcrop:
  • ChandelierCropBR {-0.02%, -0.31%, -0.02%, 99.3%, 99.9%},
  • FashionLadyCrop1 {-0.06%, -0.29%, -0.76%, 98.6%, 100.1%},
  • WaterfallForest {-0.18%, -0.66%, -0.71%, 99.5%, 100.4%},
  • WomenFootball {-0.08%, -0.29%, -0.83%, 99.6%, 100.6%},
    • Gaming_LD_HD:
  • DOTA2s360 {-0.15%, -1.77%, -1.81%, 99.4%, 99.1% },
  • GTAVs090 {0.00%, -1.51%, -0.65%, 99.2%, 100.1%},
  • Minecraft {-0.13%, -4.25%, -2.62%, 99.9%, 99.4%},
    • UGC_RA_HD:
  • Camellia {-0.08%, -0.61%, -0.61%, 100.5%, 99.6% }.

Two tests: For both CCLM and CCCM, or for CCLM only. The latter has less gain. For CCCM, also a new model is proposed.

This gain is in some sequences quite significant for chroma, and also for some cases artefacts are observed.

It was agreed to investigate this in an EE. The benefit of the additional CCCM model shall be investigated separately. Instead of using individual parameter files per sequence, an encoder-side method to detect the chroma position of a given sequence would be desirable.

Cited:
JVET-AM0312 X
ARCHIVE :
JVET-AM0196-rev.pdf JVET-AM0196-v4.docx JVET-AM0196-v4_clean.docx TestA_CTCcontent.xlsm TestAB_draftCfEconent.xlsm TestB_CTCcontent.xlsm
Decisions
It was agreed to investigate this in an EE. The benefit of the additional CCCM model shall be investigated separately. Instead of using individual parameter files per sequence, an encoder-side method to detect the chroma position of a given sequence would be desirable.
Citation