< Back to blog

AVS3: a new life of the old technologies

Maxim Lobasso
Written byMaxim Lobasso

A lot of video codecs are available at this moment. The line from ITU/MPEG: AVC → HEVC → VVC. The line from OpenMedia group: VP8 → VP9 → AV1 → AV2. And now comes a line from China: AVS3-phase1 → AVS3-phase2.

What is the reason for developing different lines of video codecs? The only purpose of compression is quite plain: to produce as little bits as possible, sustaining some video quality. Of course – money! All ITU/MPEG codecs are royalty codecs, and all OpenMedia codecs are royalty-free. Nice, we can select a codec for implementation based on its cost/efficiency criteria! (please keep in mind that the implementation cost means not only royalty) So, what about AVS3? Maybe it is royalty-free? No. Maybe super-efficient? Of course not! What is the reason for developing and implementing a new line of video codecs? I think it is simplicity. For example, only I420 color space is allowed in AVS3. I422 and I444 are good progress, but do those have any wide use?! No, I420 is still the most used format in all applications. Or picture splitting: slices, tiles, subpictures... it is a good set for parallel execution, but do you use all of them? Maybe slices (or patches, in translation from Chinese :)) are enough for these purposes... and so on at each stage of the encoding process. AVS3 standard is used as the only really effective and the most used algorithms and technologies.

AVS3-phase1 have coding-transform unit (CTU) splitting modes like HEVC and use the most efficient additional algorithms set for Intra coding:

  • Intra Prediction Filter (IPF)
  • Two-Step cross Component Prediction Mode (TSCPM)
  • Intra Dual-Tree (DT) splitting

and other algorithms set for inter coding:

  • Adaptive Motion Vector Resolution (AMVR)
  • History-based MV Prediction (HMVP)
  • Extended AMVR (EMVR)
  • Ultra MV Expression (UMVE) scheme
  • Affine motion prediction
  • Symmetric MV Difference (SMVD)

and for filtering:

  • Position Based Transform (PBT)
  • Loop Filter (LF)
  • Sample Adaptive Offset (SAO)
  • Adaptive Leveling Filter (ALF)

It allows AVS3-phase1 standard to keep its efficiency above HEVC and not to be significantly slower.

Pic.1 Heat map for AVS3 bitstreams

AVS3-phase2 currently uses a far more expanded algorithms set:

  • Extended Intra Prediction Mode (EIPM)
  • Enhanced TSCPM (ETSCPM)
  • Implicit Selection of Transforms (IST)
  • Multiple Intra Prediction Filter (MIPF)
  • Prediction from Multiple Cross-components (PMC)
  • Intra Prediction Filter for Chroma (IPF Chroma)
  • Improved Intra Prediction (IIP)
  • Spatial Angular Weighted Prediction (SAWP)
  • Intra Block Copy (IBC)
  • Frequency-based Intra Mode Coding (FIMC)
  • Implicit Selection of Transform Skip (ISTS)
  • Symmetric Intra Block Copy (SIBC)

Pic.2 Transform mode for AVS3 bitstreams

Additional inter set:

  • Sub-Block Transform (SBT)
  • Decoder side MV Refinement (DMVR)
  • Bi-directional Optical Flow (BIO)
  • Inter Prediction Filtering (InterPF)
  • Motion Vector Angular Prediction (MVAP)
  • Affine UMVE
  • Angular Weighted Prediction (AWP)
  • AWP with Motion Vector Refinement (AWP_MVR)
  • Enhanced Temporal MV Prediction (ETMVP)
  • Sub-block-based TMVP (SbTMVP)
  • Enhanced UMVE
  • Overlapped Block Motion Compensation (OBMC)
  • Bi-directional Gradient Correction (BGC)
  • Affine Secondary Prediction (ASP)
  • Inter Predictor Correction (IPC)
  • Transform Skip for inter blocks (TS Inter)
  • String Prediction (SP)
  • Equal Value String and Unit Basic Vector String (EVS_UBVS)

Transform and filtering sets:

  • Scan Region-based Coefficient Coding(SRCC)
  • Counter-based multi-probability CABAC
  • Enhanced Secondary Transform (EST)
  • Secondary Transform for intra Chroma residual (STC)
  • DeBlocking Refinement (DBR)
  • Enhanced Sample Adaptive Offset (ESAO)
  • Cross-Component Sample Adaptive Offset (CCSAO)
  • ALF enhancement (EALF)

Those instruments may give AVS3-phase2 (under development) the efficiency approaching VVC with fast execution.

Pic.3 Prediction for AVS3 bitstreams

Such simplicity permits lower resource requirements and, as a result, lower computational costs. It is not a secret that China is the world’s major manufacturer of electronics. The existing video coding standard that allows implementing lower-cost chips in video equipment compared to other standards renders significant competitive benefits. So, we can mark AVS3 standard as a standard for video equipment manufactures.

References

  • Draft text of AVS3 group standard