Yes. Notice that this shows the shoulder bends upwards, because the linear section steep enough to isn’t hit target display white. That’s one of the places (the fallback for the upwards-curving shoulder or downwards-curving toe) when instead of the set target, a 0 or 1 is used in the fallback.
Ok I played around with dt-agx today a bit, it’s kind of completely different from what I tested some time ago. Speed is ok on my Ryzen full hd desktop. I find the results somehow similar to Filmic and according to my experience the module local contrast is almost always necessary.
Edit: Well the results are similar to Filmic but better 
2 Likes
I finally was able to figure out my build issues, so I just tried this today and I have to say you’re really onto something @kofa
I look forward to playing with it more and eventually seeing it in a proper release!
4 Likes
You probably don’t need a linear section since you can just increase the power functions to extend how long the mid-section remains linear-ish. See this desmos for a simpler form if it’s helpful: PowerP Sigmoid | Desmos
In order for that function to be useful it should be applied on a log encoded image and have a power function applied overall after, otherwise the “toe” region looks pretty wonky. (Possible you are already doing this, I have not looked closely at the code).
While I’m here and since you all seem interested in “tonescale functions”, I’ll share with you my trash pile: Google Colab
3 Likes
Hello Jed,
Thanks for the feedback. Sorry about the quality of the code, will clean it up. Yes, the input is log encoded. I basically took Troy’s Python code that generates the OCIO config, and exposed the params as sliders, so yes, there is a power function at the end (using the ‘gamma’ which is used to place the pivot higher on the y axis). Currently, the pivot is not exposed via sliders, it’s set to mid-grey to be consistent with darktable’s other tone mappers, filmic rgb and sigmoid, but it’s in the plans to make it adjustable.
As for the power vs end of the linear section, I’ve observed the behaviour you describe. It’s really up to users to decide which sliders they’d like to keep - I think it’d be great if we could reduce the number of controls. (Other readers: Jed is the author of the curve, so I think we should really consider his advice.) @s7habo , if you try to re-edit the 1st image of Blender AgX in darktable (proof of concept) - #467 by s7habo with only the toe power, can you make it look the same, or do you find the toe start control is necessary? (Of course, opinions from others are just as welcome, I just remembered this particular image as being an example where the control was used.)
I just discovered yesterday that the humble ‘First time Play Raw poster’ of Rendering Greens is you. 
I’ll start reading your materials during the weekend.
3 Likes
I am at your disposal to test it 
4 Likes
There’s no mac os build for dt-agx, is there?
There is a build for arm mac that @MStraeten shared 5 days ago in this thread, not yet for intel mac. that build may be not representative, this project is evolving very fast…
1 Like
No changes since last weekend.
@s7habo and I had a brief back and forth about these sliders, and we both agreed that the toe/shoulder start sliders are more like “booster” sliders than essential sliders. I haven’t yet found an instance where they are essential per se.
They work differently than how I was initially expecting. I thought they would be basically a null operation when moved on their own, and only have an effect when the toe/shoulder power sliders are used, similar to how the shadow/highlight balance sliders in Filmic don’t do much until you add latitude and contrast.
Hi Dave @dave22152, with the new nightly build (#615) installed, the Agx module has dissappeared. could you reconfigure the windows executable, or should I revert to the earlier one which you created?
Thanks
@PD1 - to my knowledge there’s been no update to the AgX POC, so the executable in my drop box should still be valid. Let me know if you need more help
Note that AgX is still a proof of concept and in its own branch, so you won’t find it in the nightly developmental builds
1 Like
A new version with just a minor fix: target black and white should now be honoured by the ‘fallback’ toe/shoulder (used when the original curve would no longer reach those points because of low slope or not enough space in the toe/shoulder).
https://tech.kovacs-telekes.org/dt-agx/Darktable-5.1.0%2B656~gde4613e6e7-x86_64.AppImage
6 Likes
Link to Windows executable:
4 Likes
It works wonderfully.
Agx only.
High key:
Low key:
It would be good to try the simple curve @jedsmith suggested and variable gamma (not fixed to 18% gray)
2 Likes
4 Likes
here the macOS arm64 build:
darktable-5.1.0+635~g450473d104_arm64.dmg
same comments as in Blender AgX in darktable (proof of concept) - #181
2 Likes
The simple curve is the same as the current one, I think, with no linear section, and adjustable pivot. I’ll have to check (I’m not at my PC).
I’ll expose the pivot controls in the next build.
Right now, you can change the gamma, but not the ‘map from’ and ‘map to’ values (they are both set to 18%⟩.
2 Likes
Yes, that would be good to test.
Thanks Dave. Have installed as a fork, sharing the same Db. All good
1 Like