Would this now work in ART, ie AgX via OCIO???

I have not had time to read fully what @agriggio recently added to ART. It appeared that it might be OCIO support or some version of that to support some certain type of LUTS.

I was wondering if that extended to being able to use something like this

Discussed a bit here wrt video

I had another more in depth link but I have lost it for the moment…I will update the post when I find it … just curious

Yes, all of this would work with the proper CLF LUT definition. What would this be for, I have no idea though :slight_smile:

I really only stumbled on it…basically in the little scan reading I did this was a sort of newer version of filmic for Blender?? I am likely really wrong and I don’t know exactly how it would be used but it seemed like it had the ability to be a technical transform of sorts… so maybe some tone mapping??

I really don’t know… the link I am missing was far more detailed and is what got me thinking about it… I will try to dig that up…

Thanks for your feedback… Hopefully someone out there knows if this would be in any way useful…

1 Like

Hi,

If you want more information the place to go is this link

It’s a very active, exciting and instructive thread.

3 Likes

This is the link I was browsing… I need to re read it …

1 Like

My understanding of OCIO configs and LUTs is quite limited, but having spent quite some time experimenting with the AgX approach recently, probably I can comment something.

The AgX image-formation transform (i.e. transforms from scene-referred input data to display-referred, formed image, same what filmic does in darktable) takes the following approach:

  1. Convert to a working space (currently Rec.709 primaries, Display P3 is being considered as a replacement)
  2. Clip off negatives (to be replaced by more sophisticated compression)
  3. “Inset” the input values toward achromatic by a bit (this is done by a matrix transformation)
  4. Apply per-channel RGB curves
  5. Optionally apply chroma and contrast changes
  6. (Future: expand or compress toward the destination color gamut)

Currently only the per-channel RGB curve part (step 4) is formulated as a LUT in Troy’s repo, the other things are done in the OCIO config that’s distributed in the same repo. As again I’m not very familiar with how the OCIO things work, I’m not sure if the entirety of this can be formulated as a CLF LUT. If it’s possible, would be certainly worth trying.

As for AgX itself, the “inset” part is what really sets it apart from the existing image-formation methods. It allows handling bright, saturated colors much easier.

Hi,
here’s the equivalent CLF LUT. It would be good if you could confirm this is indeed equivalent – it’s totally possible I might have made some mistake…
AgX.clf.gz (18.7 KB)
EDIT TO ADD: this LUT is for use in ART, i.e. it contains conversion from/to ACES 2065-1 besides the steps you outlined above – which can all be expressed as CLF operations (again, mistakes if any are mine)

3 Likes

Really appreciate the breakdown. I was curious and this helps formulate the overall usage… thx

Great, I’ll have a look tomorrow. Would be also great to see how this was generated (to better be able to review it, as well as to learn something new)

Just gunzip it, it’s xml. I manually translated the various steps you outlined above (as seen in the OCIO config file) to the equivalent CLF nodes. Specifically, the CLF file contains 9 process nodes, which do the following:

  1. Matrix node from ACES 2065-1 to XYZ
  2. Matrix node from XYZ to linear Rec.709
  3. Range node to clamp negatives
  4. Matrix node doing the “inset” transformation above
  5. Log node to convert from linear to log2
  6. Range node to remap the log range [-12.47393, 4.026060] to [0,1]
  7. the per-channel curve as a LUT1D node, obtained with ociomakeclf from the original in the github repo
  8. Exponent node to undo the 2.2 gamma
  9. Matrix node from linear Rec.709 to ACES 2065-1

HTH

1 Like

Is the ability to use this now baked in to ART or does it need to be compiled to access it. I noted somewhere something about a flag to set for compilation. I have been lazy for some time now as a while back I couldn’t get ART to build on Windows… Would it be in the latest windows nightly binaries??

I don’t know about the nightly binaries, but it’s definitely in the latest “official” version. (And yes, it needs to be enabled at compile time).

HTH

And you access it in Film Simulations???

Either film simulation (which is what I would use for an ODT/DRT lut like this one) or colour/tone correction.

1 Like

Based on this description, looking at the CLF file and a brief test, it seems to work. :slight_smile: I haven’t yet setup the original config in Blender so didn’t do an exact conversion yet but qualitatively it matches what I’ve seen in my own experiments and some done by others.

Image from GitHub - sobotka/Testing_Imagery: Colourimetric Test Imagery, converted first from EXR to 32-bit floating point TIFF in darktable.

Neutral profile as a starting point (so that no tone-mapping was applied). Just loaded the LUT and this was the result. It is as expected. Well done :+1: Should be something quite interesting to experiment with for the folks here.

1 Like

Thanks for checking!
Fwiw, qualitatively the output is a bit too flat and desaturated for my taste, but you can probably fix that with a bit of “look” tweaks.
I prefer the current ACESnext candidate A out of the box, or even better the RED “medium contrast, medium rolloff” one, as it is closer to a finished image. But these are just personal preferences…

1 Like

The ”AgX Base” look is meant to be that way. Eary has been providing updated AgX configs here as the experimentation progresses: AgX/config.ocio at main · EaryChow/AgX · GitHub

There are several creative looks but also the ”AgX Chroma Laden” look which adds a saturation increase on top of the AgX Base look. That should be more like a ”finished image” type of starting point. But yes, personal preferences indeed.

2 Likes

Great, thanks! I’ll see if I can convert those to clf as well. I skimmed through the thread linked above but I didn’t have the energy to read all the 900+ posts in it… :slight_smile:

2 Likes

Yeah, no need. Eary’s most recent config contains also a spherical compression (done before the image formation) but that is still the topic of ongoing experimentation and it’s not yet settled on.

Here’s the “chroma laden” version of AgX as a CLF. This time I used a different approach to create the LUT, leveraging ociobakelut plus some log shaper at the beginning to ensure that it is applicable to a suitably large dynamic range. It looks ok to me (in the sense that the contrast is similar to AgX but it’s more colourful), but independent checking would be good.
You can use it in ART either by uncompressing it, or by changing the extension to .clfz.

HTH
AgX_Chroma_Laden.clf.gz (1.3 MB)

4 Likes