An AI Infused Enhancement To LCEVC Codec

  • Thread starter bhobba
  • Start date
  • Tags
    Ai
In summary, the AI-infused enhancement to LCEVC codec is a promising development that aims to improve video compression and delivery. By incorporating artificial intelligence into the existing LCEVC codec, this technology can intelligently determine the optimal encoding parameters for each individual frame, resulting in better quality and reduced bandwidth usage. This enhancement has the potential to revolutionize video streaming and enable smoother playback on various devices with varying network conditions.
  • #1
10,824
3,690
In the insights article, I did on how 8k video can be distributed at internet speeds we have usually access to these days, I mentioned LCEVC:



For more details see:
https://ieeexplore.ieee.org/document/9795094

It is a very effective technique with older codecs, but with newer codecs like VVC the gains are only small, with not using it often producing better results:
https://ieeexplore.ieee.org/iel7/76/4358651/10019298.pdf

As mentioned in previous posts, since LCEVC was standardised, AI has made significant gains. An AI varient of LCEVC will be shown to outperform LCEVC.

It uses an auto-encoder called CAESR:
https://arxiv.org/pdf/2202.00416

This is combined with an AI-infused varient of LCEVC:
https://www.tnt.uni-hannover.de/pap...ding_with_Spatial_and_Temporal_Prediction.pdf.

It outperforms LCEVC, except for a multilayer configuration.

Again, this post has introduced a number of new ideas the reader may not be familiar with. It will take a bit of time to assimilate, but already it is seen that AI outperforms single-layer LCEVC.

The next post will show how the scheme can be enhanced using the newer super-resolution techniques from a previous post and enhancements to EVC Baseline Inter coding.
 
  • Like
Likes Greg Bernhardt
Computer science news on Phys.org

Similar threads

Replies
8
Views
4K
Replies
2
Views
1K
Replies
5
Views
1K
Replies
1
Views
1K
Back
Top