Weird color fringing on highlights when processing with darktable 3.0

[DefectiveMonk]

I'm getting weird color fringing on bright reflections when processing my RAW images in darktable 3.0. In this particular image you can see it in the reflections of the moving water around the rocks. It's a pinkish/purplish color much like you might find in chromatic aberration but turning on CA correction doesn't help. I've had other images where the fringing was a cyan color.

Is this a problem with darktable, my camera, or the way I'm processing the image? I can't figure out how to attach the .ORF file but here's an un-processed JPG from it.

Subscribe to see EXIF info for this image (if available)

 

[0000]

Uploading a RAW file to some kind of shared storage would be the way to tell for sure what's going on, but a 100% crop of an affected area in the frame might allow for some good guesses... maybe with/without the basecurve module enabled.

EDIT: Oh, and a quick thing to try would be changing the chroma preservation mode in the basecurve (or filmic RGB) module to "none" or "luminance" and see if that helps. The other modes can sometimes create artifacts on clipped areas.

 

[DefectiveMonk]

As this is a spot I visit regularly I got a better example today. As you can see, there's a weird cyan cast/outline to the highlights in the upper left corner and a pink/magenta cast/outline to the ones in the bottom right.
Original ORF here: P4082918.ORF
100% crop here:

Subscribe to see EXIF info for this image (if available)

ColoredHighlightsCrop by Defective Monk, on Flickr

 

[0000]

Well... that's longitudinal lens CA... it's cyan on the far side of the focal plane, and magenta on the close side. The reason it's so strong in general is that it's at the brightness of the speculars, which are actually far, far brighter than the white level of the image. The reason it's more visible in the default darktable rendering than the camera preview image is because tone curve being applied by darktable isn't as strong. With a similar tone curve applied, they look similar, with more of the color pushed into the "almost white" range:

OOC crop

Subscribe to see EXIF info for this image (if available)

darktable crop

Subscribe to see EXIF info for this image (if available)

You should be able to save that second image and load it in darktable as if it were an *.xmp sidecar file, but it's possible this won't work due to version mismatch (I'm using a development build from a little after the 3.0.1 release)... if that's the case, I'll find another way to show you the other stuff I did. Mostly the defringe and hilight reconstruction modules... you could take it further, but would probably want to use parametric masking to avoid desaturating other stuff.

If you want to avoid this in the first place, stop that lens down!

 

[DefectiveMonk]

And today I learned something new! (Several somethings actually)! Thank you. I'll see what I can do do pull that second image in.

 

[0000]

(Several somethings actually)

Yeah, sorry, it's hard to guess which of all those things might need more explanation... it's easier if you just tell me which ones I didn't explain well enough.

Some relevant links to the Fine Manual:
https://darktable.gitlab.io/doc/en/correction_group.html#defringe
https://darktable.gitlab.io/doc/en/modules.html#highlight_reconstruction

WRT the highlight reconstruction module, this is one of the odd cases where you may want to lower the clipping threshold, since you aren't actually trying to restore color to the clipped area, but rather to consider more of the area around the clipped area as "not valid color information"... you just need to be careful about the effect on other highlights in the image, but this is an example of a case where that's not much of an issue.... if there was a bright sky with cloud detail, it would be another story...

 

[DefectiveMonk]

Thank you for the offer of additional information but, as long as someone points me in the right direction I don't mind RTFM. I've been using open-source for 20 years and that's just how it goes. I do appreciate the help. My guess is that Lightroom either handles theses items differently or has different defaults as it was something new when I switched to darktable.

Thank you again.

 

[0000]

My guess is that Lightroom either handles theses items differently or has different defaults as it was something new when I switched to darktable.

Well, if nothing else, AFAIK Lightroom by default gives something much closer to the camera tone curve, and that alone makes a big difference in how saturated those areas appear... darktable has very limited facilities for automatically matching camera output, and from what I've seen this particularly applies to 16MP and later m4/3 cameras. For the comparison above, I used a basecurve made to approximate the "natural" setting on my E-PL7.

If you still have a working copy of Lightroom, I'd be curious to see a crop similar to the ones above showing its default rendering... but if you don't feel like it, no worries.

 

[DefectiveMonk]

If you still have a working copy of Lightroom, I'd be curious to see a crop similar to the ones above showing its default rendering... but if you don't feel like it, no worries.

Nope. Saving my $10 a month.

 

[0000]

As a heads up, there are some interesting things in the works for the filmic module related to how the transitions from clipped areas are treated, using slight bloom in a way similar to what I tried to do above... should be interesting, hopefully it will make it into darktable 3.2... https://discuss.pixls.us/t/filmic-v4-on-the-way/17727