JVET-AQ0196 AHG9: Loss concealment purpose for NNPFC SEI message [S. Mate, J. Boyce, M. M. Hannuksela (Nokia)] [late]
Chaired by S. Deshpande during 14:50-15:20 on 8 July 2026.
This contribution proposes extension for the NNPFC SEI message to support loss concealment. The proposal includes the following:
- Addition of a new purpose for loss concealment to NNPFC SEI message.
- Addition of decoder activation criteria for decoder side determination of criteria to trigger loss concealment processing.
It was suggested by the proponents that aspect 1 could be considered on its own for inclusion. However aspect 2 builds on top of aspect 1 and needs to be considered together.
The error concealment is proposed to be done as a post processing. It was suggested that most decoders should be able to do the error concealment (rather than as a post processing). It was asked if any example exists where decoder can’t do error concealment and a post processing needs to do it.
It was asked if decoder sees this SEI should it not do error concealment, if supports doing it.
It was asked how error concealment will work as proposed in low delay cases.
It was asked how the proposed method aspect 2 works for slice loss. There are two types defined for this.
So far the NNPF design philosophy is NNPFA activates usage of NNPF and NNPFC does not. This proposed approach requires NNPFC to activate the loss concealment mechanism on the decoder side, which was commented to be a big change to our architecture.
It was commented that so far NNPF design has been for post processing and now this is changing it to do an operation which is more decoder related.
It was commented by one participant that the error concealment area is less important as network conditions get better. It is asserted that frame freeze may be better than error concealment in some cases.
It was asked why NNPFC needs to do the activation and not NNPFA. The proponents answered that since you don’t know which picture(s) are lost, you don’t know which pictures need to have NNPFA be sent for.
It was asked if POC gap is known by the post processing entity.
Encoder optimization information (EOI) SEI message (1)
Contributions in this area were discussed during 1120–1135 on Wednesday 8 July 2026 (chaired by J. Boyce).