[FFmpeg-devel] [PATCH v2 2/7] lavutil: add dolby vision related header

[Jan Ekström]

On Sat, Apr 11, 2020 at 5:49 PM Carl Eugen Hoyos <ceffmpeg at gmail.com> wrote:
>
> Am Sa., 11. Apr. 2020 um 16:08 Uhr schrieb Jun Zhao <mypopydev at gmail.com>:
> >
> > From: vacinfang <vacinfang at tencent.com>
> >
> > add dolby vision related header
>
> Apart from the name of the header file:
> Could you explain why it is needed?
>
> Is the Dolby Vision metadata not a black box that libavformat
> has to provide to interested decoders?
>

There's usually two different things:

1. Metadata in the file confirming that this is indeed DoVi. This is
what is being parsed. Otherwise the player application has no idea how
to handle DoVi (many of the profiles are actually just re-brands of
already standardized setups, while profiles 5 and I think 7 are the
more interesting ones).
2. The kind of "black box" (RPU) metadata within NAL units 62. That is
then what (for now) would be passed to a special metadata reader to
enable correct rendering of f.ex. profile 5 clips.

This patch set enables noting whether a file has the general metadata
from 1. I have seen implementations of 2. around, and we might have a
patch set for it at some point in the near future as we already have
open source code to render the DoVi data w/o the dynamic metadata of
those NAL unit 62 entries (see
https://code.videolan.org/videolan/libplacebo/-/commit/f850fa2839f9b679092e721068a57b0404608bdc
).

So to reiterate:
- The container metadata lets a player figure out if the AVStream is
flagged as DoVi and in which profile/level/does it have base|extension
layer contained within it.
- The NAL unit 62 metadata (RPUs) then enable correct rendering in
case of a profile that is not backwards compatible with standard color
spaces.

Best regards,
Jan