[FFmpeg-devel] [PATCH 1/3] avutil: Add Dolby Vision RPU side data type

[Derek Buitenhuis]

On 10/22/2021 11:02 PM, Jan Ekström wrote:
> DoVi Profile 5 and 7 rendering requires the data within these RPU
> entities, AFAIK (as it contains various values to correctly configure
> the custom color space). So software such as libplacebo where haasn
> has made efforts towards implementing the non-standard color space
> utilized would benefit from having the contents.
> 
> So while there clearly are two separate use cases, I would note that
> saying "nothing would see any use from the bytes within these
> structures" is maybe a bit disingenuous.

Do any of thse actually work or is it just gusswork based on reverse
engineering and patents (0% chance I am readng any patent or IPR). We
could rename this or DOVI_PROFILE8 or RPU_BUFFER or something if you
prefer.

> The two use cases being:
> 1. Just re-encoding the content. The contents of these RPU buffers
> being unrelated and only passthrough is necessary.

Yes, supported today by e.g. x265, which I imagine is the most common usecase.

> 2. Trying to gather the information of these metadata units regarding
> interpretation of the received image.

To what end is a bit iffy - nothing can make much use of 99% of it.

> Both are clearly valid use cases, and we should IMHO support both, at
> least on the AVFrame level. Not that I mean that you should implement
> the parsing of these metadata units, since it is clear you want to
> keep away from that part of the work :) .

Indeed, I was hoping this patchset would not be bogged down with "yeah
implement this undocumented proprietary parser from RE or patent instead"
which I obviously cannot do (dayjob, etc.) - and I think the value is ...
eh ... at best.

I'm not particularily fond of the idea of indefinitely killing this
functionality while waiting for someone to take up the parsing.

>> - and it is a legal minefield
>> which I am not going to touch.)
> 
> While with this I agree with completely. Or at least given the D
> company's track record :P . That said, just parsing the metadata units
> is probably much less of a swamp than actually trying to utilize the
> metadata to properly render Profile 5 or 7 video.

Which is why parsing it is iffy value given the risk.

- Derek