Oh! Then I would say it’s very impressive.
Clouds recovery would be even more interesting.
Results from the latest modifications ~ 1 stop of highlight synthesis on clouds.
Thanks to @garagecoder for his local boxfilter in GMIC 
Edit: Previous example with new modifications. Highlight synthesis and my custom colour propagation.
2 Likes
Had some fun with trying my highlight synthesis with images from other threads.
Top; RT. Bottom; Mine.
From:
5 Likes
Excellent results!
1 Like
I’d be so thrilled to see this in Dark Table !!
1 Like
Here you go. Please bear in mind that these images are not in a calibrated colour space. That is, I just use the RAW RGB values. EG The problem with over-saturation of the blues would still probably happen with normal colour management.
6 Likes
Thank you! Amazing. Simply amazing.
I wish this was implemented in Darktable. Over-saturation could be solved by the color calibration module.
You can try it now:
GIMP → Filters → G’MIC-Qt → Testing → Iain Fergusson → Iain Highlight Synthesis
1 Like
@Iain and i have been working heavy on a darktable implementation of his ideas, many things had to be implemented in a slightly different way or required new stuff in dt. It’s already usable with good performance (on my 8core E-2288G the algo takes <100msec on the selected image), some issues remaining but we both are confident we find a good solution.
Here is a screenshot, no filmic applied so no interference from there, slightly dark for better view of the highlights, just default color management.
12 Likes
I’m telling you this is the best darktable news I’ve heard in a long time. 
So looking forward! Handling of blown out highlights has (for me) long been the achilles heel of dt. “reconstruct in LCh” often does too little and “reconstruct color” has a very high chance of really nasty artifacts (such as pink striping/patterns). How common are strange artifacts like those with @Iain 's color propagation method?
Lastly, is there any chance whatsoever of this ending up in the 3.8 release? 
Super nice to see this heading towards a dt integration! Any branch or PR up that one can try out?
No. 3.8 is already in feature freeze
There’s no chance of those weird pattern artefacts. Other kinds of artefacts, well, that’s different. 
But seriously, all highlight reconstruction is guesswork. The approach we are taking, splits clipped areas into patches and tries to guess the correct colour for that patch. The most obvious artefact is when the guessed colour is completely wrong.
A large patch of the wrong colour should be more easily corrected with other tools than weird patterns and smudged colour.
I am still hoping an RT dev takes up the effort as well.
Nope - not yet
This stuff is overall a difficult beast. We are still sorting out “how to do” in the algorithm and have to keep performance controlled.
Yes - there will certainly be a pr later, a wild guess would be soon after 3.8 release.
And yes - personally i already use this for my daily photo work
6 Likes
OK, thanks!
Arghhh, that’s teasing! 
You could open a draft PR on GitHub so people could find the bugs more quickly 
Atm it’s not about bugs but concept
1 Like