JVET-AQ0188 Subpicture support in HEVC [A. M. Tourapis, D. Podborski, J. Kim, S. Paluri (Apple)]

Chaired by S. Deshpande during 14:30-15:20 on 12 July 2026.

This contribution is a resubmission of JVET-AP0263. It updates the subpicture support proposal for HEVC first presented in JVET-AI0256 and further developed in JVET-AJ0219. The updated design introduces a revised subpicture_info SEI message with a cleaner architecture based on the concept of composited pictures. The SEI defines an atlas grid for spatial layout, maps HEVC layers to subpicture regions, and associates each layer with a composited picture type that determines its rendering method and purpose. Per-layer properties such as view identification, auxiliary type, resolution, and colour description are obtained from the existing VPS, SPS, and VUI, avoiding duplication. Decoded layer content is scaled to fit the subpicture region, decoupling coded resolution from atlas geometry. In this version, an optional modification of the Atlas Definition is additionally described, in which the subpictures (atlas segments) are defined directly by their location and size, in a single step, instead of through the region grid; this is offered should the group consider the region-grouping design too complex.

Slides to be uploaded in a revision.

The intent is not to replicate VVC subpicture functionality, but to have a much more generic version than VVC functionality.

Full atlas can be an output picture or a container for output pictures.

Atlas allows understanding the relation of layers together.

It was commented that this is beyond the scope of coding specification (like some system level information).

One atlas corresponds to one time instance.

JVET-AQ0136 concept of target picture is similar.

What is the use case to allow overlapping: an example is picture-in-picture.

Priority id metadata tells how to handle overlap and order of processing.

The proposal is not HEVC specific and can also apply to a future codec (VVC already has subpictures).

Atlas can also have holes (empty regions).

Display overlay SEI in TuC is most related but does not have all the functionality proposed in this SEI.

Different subpictures can have different number of associated layers.

Combined discussion for JVET-AQ0136 and JVET-AQ0188:

-Do we want this functionality in SEI:

-Subpicture functionality: has more non-proponent support

-Composition functionality: some non-proponents felt this should be done at system level.

-If so, should it be a new SEI or extension of some existing SEI: It was suggested to check if PRI SEI can be extended to support the subpicture functionality, including supporting multiple target pictures.

Regarding the aspect of supporting multiple target pictures in PRI: there was some objection to add that, and rather only support subpicture functionality.

It was also commented that an complicated SEI to support just the required functionality may not be best.

It was commented that PRI already has some concept of composition, although not a full fledged composition functionality.

It was noted that only one of the two proposals have multiple target pictures.

It was commented that a possible drawback of PRI is that there are N regions and each region is associated with only one layer. And so if PRI is used for this functionality this aspect should be considered/ modified.

Further study based on above direction.