AI for Encoding Coming in Different Phases

What They Say

Thierry FautierThe 8K Association published an article that quoted Thierry Fautier, VP of Strategy at Harmonic who talked about how AI and machine learning (ML) could be used to optimise content for existing codecs including AVC, HEVC, AV1 and AVS3. He expects a second phase to focus on VVC and AV2.

Harmonic already uses AI to assess video quality in real-time and support ‘Content-aware-encoding’ (CAE) and the firm now has over 100 deployments globally. The use of CAE can reduce the bit rate needed for 8K. The system uses existing encoder hardware.

Developments are in dynamic resolution encoding (where per title/per scene encoding is applied) and Harmonic is focusing on live techniques rather than multi-pass as might be used for streamed content. The firm is also working on dynamic frame rate encoding to drop the frame rate when high frame rates are not needed. As much as 30% of the compute load for encoding can be saved and Fautier said that AI is really helping this to develop.

The third area is using layering, where a base stream is paired with an enhancement layer for 8K. An advantage of this approach is that they can be delivered separately – e.g. the base layer by traditional broadcasting and the enhancement layer over IP. (A New Compression Paradigm)

What We Think

I had planned a deeper dive into this topic, so delayed the article, but it will have to wait for now, so I thought I had better publish it before it was ‘olds’ instead of ‘news’. (BR)