[FFmpeg-user] Dolby Vision

[Ted Park]

> The colour (and luminance) which has been decoded, mapped between
> colour-spaces, and and re-encoded is not correct. If you have a Dolby
> Vision compatible TV you can see the difference very clearly. Do you
> need me to point a camera at my TV to demonstrate :-)

I understand that the colors are different, I was saying it is being decoded the way an hevc decoder should. I didn’t realize it would look “wrong” as in a completely different transfer function used on it until I actually downloaded the file… That file only has a single stream, with no enhancement layer for the dolby HDR metadata. It just has the base layer with the “rpu” with all of the info your TV’s decoder can use, but are simply discarded, as far as I can tell. This is the correct behavior though, and the hevc stream is being decoded “correctly,” but in this case it is not sdr/hdr compliant so the base layer decoded the way it’s supposed to be doesn’t look right. I can’t help but think any fix to this would have to rely on such heuristics to convert between color models, but even before the transfer functions, dolby uses different primaries, so the best we could hope for is an approximation. This is just the profile in the reference file you linked, in practice most distribution paths wouldn’t have only the single layer non-compatible dolby vision stream, I assume this was meant to be demo material on floor display TV units or something.