Thursday, May 28, 2020
9:30am - 10:30am PT (Los Angeles time / UTC-4:30pm)
Please join us for the next meeting of this virtual working group (VWG). Future meeting dates for this month include:
TBD
Dropbox Paper link for this group:
We will be using the same GoToMeeting url and phone numbers as in previous groups.
You may join via computer/smartphone (preferred) which will allow you to see any presentations or documents that are shared or you can join using a telephone which will be an audio only experience.
Please note that meetings are recorded and transcribed and open to the public. By participating you are agreeing to the ACESCentral Virtual Working Group Participation Guidelines
Audio Only
You can also dial in using your phone.
Dial the closest number to your location and then follow the prompts to enter the access code. United States: +1 (669) 224-3319 Access Code: 241-798-885
A question was put to the group of whether to focus more on @jedsmith’s RGB based approach or also to continue to pursue other types of algorithms. The general consensus was to focus on the Jed’s method with any other investigations only taking place in the background.
People like the simplicity of RGB method, staying in that colorspace is ideal
@daniele pointed out that this method is closer to film emulation / real world colors in appearance, could help with adoption
What should the confidence gamut be? The area that encompasses the MacBeth 24 chart, the ColorChecker SG chart with the Munsell Colors or Rec.709? The definition of the confidence gamut would also need to depend on the target gamut to allow for a boundary region for the compression. The group will be focusing on AP1 as the target gamut.
Should we be mapping from infinity? Daniele points out that IDTs are the worst they’ll ever be right now, and should only get better - we hope this is true!
@joseph pointed out that we also might eventually need gamut mapping for cameras that might satisfy the Luther-Ives condition - from spectral locus to AP1
Has thresholds for individual RGB channels instead of a single threshold been tested? No, it’s an interesting idea.
@matthias.scharfenber brought up that any rolloff proposed for shadows to compensate for changes in noise breaks any idea of exposure invariance.
Jed has a version for testing the effect of shadow roll off.
@LarsB brought up the issue with doing per-channel compression causing possibly single-colored noise in the shadows
We should maybe create a test for this case with achromatic noise
If we are using a gamut agnostic, what about invertibility? Should we be inverting colors at the boundary back to infinity? Where / when / should this happen?
Will this algorithm’s inverse ever be applied on non-gamut mapped data?
How do we evaluate our algorithm under display referred conditions?
Evaluate against other colorspaces / output transforms in a non-ACES spaces
The idea was suggested of a simpler display transform for testing, at least without the “RRT sweeteners”.
Not all the imagery that enters into the system does it via a vendor blessed IDT, e.g. RAW to ACES, etc…
The IDTs will hopefully get better but we reasonably cannot build a solid system while thinking it will be the case or we need to impose absolute requirements on IDTs themselves, but as of now, AFAIK, there are not.
Mapping from infinity is certainly not practical and we are bound to the reality of floating-point number representation (and storage) with maximum representable values being:
Half16: 65504
Float32: 3.40282E+38
With that in mind and given that (to my dismay) inter-exchange must be done with Half16 as per the ACES container specification, it puts 65504 as a reasonable absolute limit: Under the current specification, nobody will ever receive or send files with ACES relative exposure values over 65504.
From Spectral Locus to the intersection of AP1 with the Spectral Locus. We need to consider the non-physically-realisable values beyond the Spectral Locus that AP1 can represent.
In Set Theory, the union of AP1 and Spectral Locus would be the set of elements belong to either AP1 and Spectral Locus or both of them. We want the set of elements that belong only to both of them, i.e. the intersection, i.e. AP1 ∩ Spectral Locus.
Exactly. The above read to me like “in AP1, but outside the spectral locus” which is not in the intersection. But maybe I’m misunderstanding what you’re getting at.
