Path Forward?

Discussions - ACESNext VWG – Gamut Compression/Mapping
1

Zone of Confidence

The VWG requires a settling on a reference zone of confidence. Mandatory to establish the volume hull of the agreed upon zone of confidence to implement nonlinear transitions.

Zone of Confidence Questions

  • The generalized reflectance chromaticities seem like a logical fit here? A direct naive gamut map from camera to target risks completely imbalancing the overall image “intention” by lopsidedly stressing the image as “oversaturated by default.” Reflectance charts seem like a fair trade off here for a region to maintain.
  • Zone of confidence plays a role in the luminance targets. An example would be non-Abney magenta shift, as a direct byproduct of gamut clipping. For example, taking typical wider gamuts to sRGB output will lift the luminance to the respective sRGB value, and exacerbating the deep indigos to more-luminous purples.

Hue / Saturation Model

The details of the reference working model needs to be established. While something like JzAzBz has acceptable results for the blue - purple and red - yellow hue skew pinch points, it is also problematic from the EV invariance need. It may be plausible to craft a parametric EV invariant nonlinear xy model, using polar coordinates.

  1. Establish hull of Zone of Fixation. Direct linear xy paths in the volume up to the hull edges.
  2. Apply a quadratic bezier curve to certain hue lines that are the most problematic, such as blue purple and red yellow. This would form an infinite plane along Y. A quadratic in xy would meet @carolalynn and @matthias.scharfenber’s to be IDT agnostic.
  3. Allow a simple tension to control the curvature, and extend from the white point outwards, for tunability during reference design work.
  4. Interpolate linearly from the radiating nonlinear hue constancy. Remain linear at some polar angle distance after such lines? Establish four lines and sample between?

Hue / Saturation Model Questions

  • Are there other options here that maintain EV invariance? Does holding naive Y at a precise value and using xy provide the sole path forward or are there other models?

Stages

It is evident that the discussions return to the idea of staged gamut annealing.

  1. Beyond AP0 to AP0
  2. AP0 to AP1
  3. AP1 to ODT

Stages Questions

  • Given that AP0 represents non-data, should the golden baseline image zone be AP1? IE: Should the image appear optimal at AP1, forming the gold standard for the ODT gamut volume compressions? BT.2020 is currently a reach in display technology, and would form a reasonably future friendly target for a golden image standard point to rally around. Until everyone moves into spectral… :wink:

Testbed Imagery

It was mentioned that there was a test bed of imagery. It might be good to have some of the worst offenders in there, and make it available? Where is it currently?

Testbed Imagery Questions

Thought I’d throw this out there for discussion. With some of the above currently-fluid facets in place, it would make proofs of principle much easier to execute via Python, etc.

1 Like
2

Hi Troy,

Thanks for this. I’m still working my way through some of the details (there’s a lot in here) but here are a couple answers:

Our sample image set lives here - the image set lives in dropbox. Feel free to add any images (that we have the rights to) there!

As far as stages go - yes, as outlined in @daniele’s document from one of our first meetings in January, there are different stages at which gamut mapping occurs, for different reasons, and the toolsets for these mappings may be different. As stated in our progress report, we are trying to set our scope to the scene-referred ACES portion of the chain only. Now, what that means for how we handle the ‘non-data’ outside of AP0 from IDTs has not yet been decided, though for invertibility’s sake, treating them as valid values so that they can come back at the end of the pipe for transfer/archival - maintaining 2065-1 - is important.

As this is an ACES group, our main focus, yes, is to make the gamut mapping work for ACES - but as @matthias.scharfenber and I have been saying, if we can, keeping it gamut agnostic (in scene-referred world) would mean trying to stay away from optimizing for a certain gamut (on either end) for as long as we can. I know that doesn’t really answer your questions - but as stated in the last meeting, and you have written above, we’d like to start with the ‘zone of trust’ and also decide in what dimension/model/space we’re mapping in.

1 Like