Re: Conversions for HDR Canvas Specification

Hi Jeff et al.,

I see two complementary objectives:

- specify mathematical expressions exactly, e.g. 4/3*pi*r^3, so that
the can be computed to arbitrary precision
- provide worked examples at a common precision, e.g. 4.188790205*r^3
at 10 significant figures for float computation, so that interop can
be achieved

Makes sense?

Best,

-- Pierre

On Thu, Jun 17, 2021 at 3:49 PM Jeff Gilbert <jgilbert@mozilla.com> wrote:
>
> I can't speak for others, but I don't want very precise *results* of
> calculations per se, but rather I prefer to know the calculation
> methods themselves, and their precise inputs. It's difficult to reason
> about results without a clear path for how we got there.
>
> Eventually I do need the numbers to calculate with, but I got tired of
> trying to figure out if matrices I found in documentation were right
> for my particular usecases, and made this:
> https://jdashg.github.io/misc/colors/from-coeffs.html (the
> calculations are fairly legible in the JS source)
>
> Ideally we can give names to constructs, instead of, for example, "use
> the following very precise matrix given to 10 sig-figs...". If there's
> a constant that's 18.0/19.0, it's nice to know that, and not just see
> 0.9473684210526315.
>
> I've found this to be a huge ongoing problem and source of confusion
> in colorspace work, but that naming things and showing-my-work goes a
> long way to helping.
>
> On Thu, Jun 17, 2021 at 3:30 PM Pierre-Anthony Lemieux <pal@sandflow.com> wrote:
> >
> > > Please, no long numbers!
> >
> > I do not understand this blanket statement.
> >
> > The accuracy of individual constants within an expression should be
> > sufficient to achieve the desired accuracy for the entire expression.
> >
> > For example, it would make no sense to specify π = 3.14 in f(x) = π·d
> > if f(x) is intended to compute the circumference of a circle to an
> > accuracy of ±0.00001.
> >
> >
> >
> >
> > On Thu, Jun 17, 2021 at 2:39 PM Lars Borg <borg@adobe.com> wrote:
> > >
> > > Please, no long numbers!
> > >
> > > And I’m not alone on this.
> > >
> > >
> > >
> > > Lars
> > >
> > >
> > >
> > > From: Chris Lilley <chris@w3.org>
> > > Date: Thursday, June 17, 2021 at 3:00 AM
> > > To: "public-colorweb@w3.org" <public-colorweb@w3.org>
> > > Subject: Re: Conversions for HDR Canvas Specification
> > > Resent-From: <public-colorweb@w3.org>
> > > Resent-Date: Thursday, June 17, 2021 at 3:00 AM
> > >
> > >
> > >
> > >
> > >
> > > On 2021-06-16 11:53, Simon Thompson - NM wrote:
> > >
> > > Hi all,
> > >
> > >
> > >
> > > I’ve added a list of suggested colour forward and reverse transforms between extended-sRGB, extended-linear-sRGB and bt2100-hlg and a suggested simple tone-mapping for displaying bt2100-hlg on an sRGB display to issue #50 - https://github.com/w3c/ColorWeb-CG/issues/50.  I’ve tried to follow the style of the example in TTMLv2 provided by Pierre.
> > >
> > >
> > >
> > > I can recalculate the matrices to a different number of significant figures if necessary – is there a consensus on accuracy levels for web operations?
> > >
> > > In general, since "space of the printed publication"is no longer a concern, I prefer to list to full precision.
> > >
> > > I have seen several problems arising from round-off errors and round-tripping, so there is no good reason to round off.
> > >
> > >
> > >
> > > Would Dolby colleagues be able to add the bt2100-pq versions to the list?  There is a PQ example in Pierre’s TTMLv2 link.
> > >
> > >
> > >
> > > Best Regards
> > >
> > >
> > >
> > > Simon
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > >
> > > Simon Thompson
> > > Senior R&D Engineer
> > >
> > >
> > >
> > > --
> > >
> > > Chris Lilley
> > >
> > > @svgeesus
> > >
> > > Technical Director @ W3C
> > >
> > > W3C Strategy Team, Core Web Design
> > >
> > > W3C Architecture & Technology Team, Core Web & Media
> >

Received on Thursday, 17 June 2021 23:26:28 UTC