Re: Next telecon agenda + review presentation

On Fri, 2 Jun 2023 10:51:39 -0700
Pierre-Anthony Lemieux <pal@sandflow.com> wrote:

> Good morning/evening,
> 
> In preparation for our next call on 6/14, please review and provide
> feedback on the following contribution from Chris Cameron:
> 
> https://docs.google.com/presentation/d/1NdrRD8fRqE6_-thfOS5DSofwsFHx0cmAoTydd-aw6_s/edit?usp=sharing&resourcekey=0-t9KRE1ckt4IFDvi0zKH-xQ
> 
> A primary agenda item of our next telecon will be to continue the
> discussion on this contribution.

Hi all,

I don't think I can make it to the telecon, so I'd like to pose some
questions here if you don't mind. These are about the presentation
linked above.

What is the "gainmap" as referred there?

I got the understanding that when you have a HDR canvas, drawing things
into it would go through tone mapping. At the same time, the
presentation explains that (for SDR workflow), color conversions are
purely colorimetric and unbounded, meaning that P3->sRGB->XYZ->P3 chain
of conversions is an identity (up to precision). Does this mean that
gamut mapping is not performed, or that it is fully reversible?

I am going to guess "not performed".

Why the inconsistency of using tone mapping but excluding gamut mapping?

If the canvas color space is intended to be a mere encoding without any
associated meaning like a specific appearance, then why not just handle
the dynamic range similarly unbounded like you have the color gamut?

Doing a simple colorimetric conversion, which is like a change of basis
in linear algebra, you allow mixing contents in a way that does not
preserve their appearance. The contents are moved into the same
colorimetric coordinate system, but they may not agree on viewer
adaptation or viewing conditions. Or, indeed, the level of graphics
white, or the HDR headroom.

If on the other hand the canvas is supposed to have a consistent color
space with a specific appearance, then you would need not only tone
mapping but also gamut mapping and chromatic adaptation. This requires
more things to be defined:
- adapted white point (probably coinciding with the color space white
  point)
- reference or graphics white level (luminance) a.k.a SDR white
- target color volume(*), including target dynamic range which defines
  the HDR headroom
- maybe more?

In defining the above luminance levels, especially if in cd/m², you are
probably committing to a specific viewing environment.

> E.g, sRGB➔PQ➔sRGB will be darker than the original.

I suppose it is due to choosing the tone mapping algorithms such that
they cause that darkening, but why would you choose your tone mapping
algorithms like that? Is this some confusion about cd/m² supposedly
being an absolute unit, or the different systems having different
reference white levels, maybe due to different assumed viewing
environments?

Luminance clipping should never happen, because PQ system is perfectly
capable of including the whole SDR, so the tone mapping should be
reversible.

I've read that ideally, the average luminance of HDR content is not
much different from SDR content, because the point of HDR is not to
make everything brighter. But if round-tripping SDR content through PQ
system causes it to be darker, then something seems off.


Thanks,
pq


(*) Sebastian has drawn really nice concept diagrams about target and
other color volumes:
https://gitlab.freedesktop.org/pq/color-and-hdr/-/tree/main/plots#primary_container_target_color_volumespy

They help make the difference between the parameters used for color
encoding from those describing the expected content, even with
unbounded encoding.

Received on Tuesday, 13 June 2023 22:14:29 UTC