Blender AgX in darktable (proof of concept)

[Donatzsky (Nis Donatzsky)]

@kofa

Please, I’m glad people find the code useful but what I have done and will do is just the mechanical stuff: exposing the controls of the curve someone else created, applying matrices someone has designed.

That’s essentially what Apple did with the iPhone

@paperdigits

So while they provide a nice test case, almost nobody will have images like that in real life & testing against real images is still important.

Here are some:

Guys I just discovered LEDs

Hi everyone. I recently discovered something called “LED”. I now own two LED appliances - one called “RGB light”, and a “Royal Blue 10W COB LED”. Here’s one of my first tests: [My favourite lenses under Royal Blue and RGB Red LED lighting] Processed with experimental program. So excited about the creative potential of LEDs! And proud to be an early adopter. (Yes I know LEDs have been around for years) EDIT: I think my first edit was a little too light? [dimmer] 5D3_9253.CR2 (23.5 MB) …

Deliberately difficult gamut colors in macro shot of hex nut

As a companion example to this topic I created, View/Modify color matrix in DCP files , I deliberately shot this extreme macro of a hex nut, 1/4" by 20, lit by incandescent bounced off a slightly yellow ceiling, with a background of pure blue (rgb 0, 0, 255) displayed on an LCD monitor. I had to export as tiff with tamed gamut, and then do curves on that tiff before export, hence me sharing two sidechain files used to get to the final jpeg shown. Have fun, and feel free to use this in any so…

Difficult orange flower

Very difficult orange flower pic. My workinprogress program, preserving tonality (but failing to preserve colourfulness, unnaturally pale): [flowersp] Rawtherapee 5.8 auto-matched curve, 4400K, very posterised: [Screenshot 2021-09-18 at 18.02.58] Lightroom, also 4400K: [Screenshot 2021-09-18 at 18.03.23] I would say lightroom’s average shade of orange looks most like the actual flower, but this is just a case of lucky clipping. 5D3_0104.CR2 (23.7 MB) This file is licensed Creative Co…

Flower Power with Roses

I took this Picture like 2 Years ago when i started photography. Like to see what others can do out of it. 20190614_RAW_Blumen_0017.nef (20.6 MB) This file is licensed [Creative Commons, By-Attribution, Share-Alike] (Creative Commons, By-Attribution, Share-Alike) This is what i have done in Darktable. 20190614_RAW_Blumen_0017_02.nef.xmp (10.0 KB) [20190614_RAW_Blumen_0017]

Red Rhododendron... why are some flowers so difficult?

I thought this plant was looking particularly good this year so grabbed the camera. Watching the in camera JPEG on the screen while still looking at the real thing showed what a poor job the camera was doing. The default DT effort was terrible too. I’ve tried myself with Darktable but looking out of the window… my efforts still look nothing like the real thing - some parts of the petals end up looking pink and the whole thing looks quite dull. The real thing is a lovely deep red… there is no pin…

[kofa (István Kovács)]

I have some of those (LED-lit and flower shots) in my collection of test images.

[priort (Todd Prior)]

There was that image I recall I think Boris used it or worked on it of a lady in this red dress lying on a set of stairs…I think it was around the filmic color discussions but that might be a nice one to try with AGX for comparisions…I will try to go back and find that one…

[Christian-B (Christian Bouhon)]

@priort

There was that image I recall I think Boris used it or worked on it of a lady in this red dress lying on a set of stairs

Hello,
I think you mean this photo,
https://www.signatureedits.com/wp-admin/admin-ajax.php?action=useyourdrive-download&account_id=103498887174941726250&id=1aUbaUq8TArNxWmjFSC-cE9uW8eUO-vVM&dl=1&listtoken=d8b65b66692c59f215d41b174d2a67af
Greetings,
Christian

[priort (Todd Prior)]

Thanks you are right…that link didn’t work for me but I scrolled and found it in the film collection I think…there were a few other good examples that could be tested that are not landscapes or concert…and yet there are some really vibrant colors…that image of the dress is image file000551, there is a good sunflower backdrop with a model and several other vibrant ones especially with models having dark skin tones and vibrant dress and makeup that would likely be good test shots to contribute to the “portrait” /human subject image type…

[kofa (István Kovács)]

@europlatus

harsh transitions and abrupt colour changes on the picture of the children:

If you disable the tonemappers, you’ll see that there are harsh transitions, as there are areas illuminated by those lamps, and others illuminated by surrounding light:

This is what sigmoid + smooth + sRGB and AgX produce on my machine:

AgX is much more desaturated, but the transitions are also less harsh, and the face is not purple.

With AgX saturation at 2:

‘Punchy’ with gamut controls:

(The matrix stuff is still not done.)

[Billal]

Agx is looking good already
How would be the matrix control design introduced ?
Would it be like sigmoid with a primaries tab, or maybe like you mentioned before it may have a seperate module at the very end with six controls ?

[Dave22152 (Dave Goldberg)]

I had played with these same images my self. Looking at a side by side of your Sigmoid and AgX “punchy” edits:

Not only are the transitions on the child’s face better controlled, but I think the overall contrast and highlight roll-off is better in the hat and scarf.

Given that we won’t see AgX in the mainline for a while, I’ve been working to see what I might do with to better control these conditions when using Sigmoid or Filmic. I’m finding I can make improvements, but it takes a bit more effort than with this POC.

[Billal]

I highly recommend to try and do exposure sweeps from 0 EV to 6 Ev or so, to see the potential of Agx as it holds and does not break visually by creating a sort of sub-gamut where a certain threshold can’t push through or create a sort of a hole in the overall image. This is the concept of additive energy, it is cohesive to our visual cognition.You can read about it in his github Wiki.

[priort (Todd Prior)]

Some time ago I did it with the blue light that you would use a second instance of CC and blend in average and start with opacity zero and work up. Here I set the red channel to 1 …it could be less but I use opacity to set the amount and then you can also try this in bypass or use the perceptual CAT and choose as in pipeline…

Here is one with only this tweak and the sigmoid smooth preset…

I’m sure a more sophisticated approach would be to use the color primaries in sigmoid with additional tweaks or to use the primaries…but again is we can avoid that or do less using the final AGX version even better…

[Dave22152 (Dave Goldberg)]

Yeah, that’s nice. I’ll have to give that a try.

[priort (Todd Prior)]

Blending in chromaticity might even be better or a different option as well… 25-35% opacity looked okay to me

[kofa (István Kovács)]

@Billal

Would it be like sigmoid with a primaries tab, or maybe like you mentioned before it may have a seperate module at the very end with six controls ?

It must be part of the module. The insetting and rotation of the primaries are needed to manage colour shifts and desaturation during the per-channel application of the tone mapping curve.

[Billal]

Super exited

[kofa (István Kovács)]

A new build:

• This one works in the output gamut, not in Rec 2020. There are still adjustments (yes, the matrix…) that I have not updated.
• You get 2 sliders to control gamut compression, both under the advanced section:
  
  
  • use input gamut compression if you observe weird colours without pushing saturation (e.g. LED lights, or simply super-saturated input colours etc.).
    
    
    
    
    
  • use output gamut compression if you pushed saturation and get artefacts.
  • note that both work on negative components only; if you get values over 1, you have to reduce brightness or back off with saturation
• I’ve added a kind of vibrance control that should affect more saturated colours not so heavily. You’ll have to tell me if it’s worth keeping.
  
  
  
  
  

As always, updated to the latest master version.
https://tech.kovacs-telekes.org/dt-agx/Darktable-5.1.0%2B766~gba2bc2f1e9-x86_64.AppImage

@Dave22152 / @priort , @MStraeten , please update the Windows / Mac binaries.
Happy Easter!

[s7habo (Boris Hajdukovic)]

It works wonderfully.

However, I wonder if we need to have input gamut compression earlier in the pixelpipe because all other color modules depend on it.

We have gamut compression in color calibration module but it does not work properly. At least I could not get as good results as with your version in Agx.

[kofa (István Kovács)]

@s7habo

However, I wonder if we need to have input gamut compression earlier in the pixelpipe because all other color modules depend on it.

The other modules work in the much wider Rec 2020 space (or whatever working space one selects); this one tries to prepare the image for display, and uses the output profile.

That said, providing it as a separate module is no big deal.

I may remove it from AgX (technically, something is needed – logs vs negatives), but the matrix insetting (which will come, I promise) may make it redundant there.
There are actually 6 parameters:

• the portion of the 0…1 range to leave alone, for each channel (currently hardcoded to 80%)
• the max ‘distance’ from achromatic to compress into the top (now: 100 - 80 = 20%) of the gamut. In my current implementation, this is set to the same value for all channels, the one you control through the slider.

The algorithms used is from Jed Smith: gamut-compress/docs/gamut-compress-algorithm.md at master · jedypod/gamut-compress · GitHub

Specifically:

GamutCompress.glsl

[kofa (István Kovács)]

Does anyone still need the old Linux AppImages? If no one reports the need to keep them, I’ll remove a few, as my hosting storage is now at 91%. (No need to report back if you don’t need them; please respond with the version I should keep if you need some of them to remain available.)

[s7habo (Boris Hajdukovic)]

@kofa

Does anyone still need the old Linux AppImages?

I would say no. I don’t think anyone would want to try all the previous versions. Maybe just leave the last two.

[Dave22152 (Dave Goldberg)]

Here’s the link to the Windows executable for the latest version:

https://www.dropbox.com/scl/fi/oyo7bua7gnsxtmt0lxxrb/darktable-5.1.0-766-gba2bc2f1e9-win64.exe?rlkey=0z2koupft69j4ov22urzlc9rg&st=nqs95tpk&dl=0