[FFmpeg-devel] [PATCH v2 0/3] avcodec: move HDR10 (MDCV/CLL) SEI handling to h2645_sei

[Jan Ekström]

On Tue, Jul 25, 2023 at 10:29 PM Jan Ekström <jeebjp at gmail.com> wrote:
>
> This allows parsing code to be re-utilized from H.264, as well as probably
> from VVC in the future.
>
> This additionally eases verification of the AVCodecContext side data patch
> set, which includes libx264 integration for HDR10 side data.
>
> Changes from v1:
> * Reordered the new h2645_sei include to correct location as per alphabetical
>   order. Thanks for Leo Izen for noticing this.
> * Cleaned up the mastering_display_metadata.h include in hevcdec, as that file
>   no longer directly handles AVContentLightMetadata or
>   AVMasteringDisplayMetadata.
>
> Notes:
> * In addition to testing with FATE, looking at the show_frames ffprobe output
>   for a UHD BD sample with each IRAP containing these SEI entries, the
>   behavior is the same as before with regards to each frame having the side
>   data persist.
>
>   There is also no change with tests I've done with a UHD BD mp4 remux as well
>   as an iphone 12 sample I had around, So I think I have not modified the HEVC
>   behavior in any great manner.
> * As for H.264, it now automatically gains support for these SEI messages. It
>   seems like its definition of CVS (coded video sequence) is from one IDR to
>   another (essentially a GOP). That said, currently the decoder resets the SEI
>   setup in decode_nal_units, which means that the side data only persist for
>   a single frame.
>

Unless there are objections, I will apply this tomorrow evening as
this is a strict improvement over the current situation where the
parsing is limited to HEVC and at least checking with ffprobe does not
show regressions for HEVC.

Jan