here’s another example file to test ‘Avoid color shift’
http://www.rawtherapee.com/shared/test_images/demosaicing_city_1.pef
1 Like
fwiw, I was very bothered by it, to the point I stopped using auto ca correction altogether. avoid colour shift made me reconsider. it’s not a simple uniform desaturation, if you have pictures of people, it might cause skin tones to be slightly off, to the point of being annoying (to me at least).
An example with and without “Avoid color shift”
Avoid color shift = ON
Avoid color shift = OFF
I’m beginning to wonder if the chroma of the usually overly saturated CA fringing is giving an artificial overall “more saturated” look to the original. This is visible in the yellow grating crop. The original looks more saturated than the one with CA removed. But in fact, the colour shifting in the central yellow region is worse when “avoid color shift” is checked. It could be our eyes getting fooled. Anyway, most pictures are too saturated and having less saturation is hardly ever worse, plus it’s easy to add more saturation later if you like.
1 Like
I can’t tell which is closer to the original without the original crop?
1 Like
CA correction=OFF
That’s a great example @TooWaBoo. Avoid color shift ON is indeed closer to the original in this case.
It’s quite remarkable how much shifting is happening here even though the amount of CA to be removed is relatively tiny. Seems too aggressive.
1 Like
Chromatic aberration changes our perception of the whole region in which it manifests itself.
Even though the hue of the flat area had not changed when enabling raw chromatic aberration correction when zoomed-in at 800%, our perception of the area when zoomed out did, and the area now looks not only desaturated bit different in hue. “Avoid color shift” returns its look to the previous state, less the chromatic aberration.
And, for comparison, lensfun:
1 Like
Indeed, but the “desaturated” CA corrected version might actually be closer to the truth/reality, without the illusion of the high-chroma fringing to distort your perception.
1 Like
Btw: When doing this tests, disable the auto matched tone curve. It will be recalculated when raw ca correction parameters changed.
Edit: scratch that. I was confused by console output which could be improved…
1 Like
I probably should mention that I begin with a neutral profile, use DCB for demosaicing, and apply no auto matched tone curve or any other adjustments when making these tests.
2 Likes
About the default value for number of iterations (2) I still think it’s a good choice because in the cases where 2 iterations are worse than 1 iteration the result is not much worse whereas in the case where 2 iterations are better than 1 iteration the result is much better. Finding a good default is not easy and often needs more time than coding…
About the ‘Avoid color shift’ being default. There have been also discussions about forcing ‘Avoid color shift’ without the possibility to disable it… Fortunately we kept it for reasons.
Imho the main drawback of ‘Avoid color shift’ is that it has to be improved (and there is room for improvement I guess)
1 Like
Thanks for sharing so much invaluable information and these further thoughts.
I do agree that 2 iterations in most cases are better than 1. This is true for a number of my other images. The autumn leaves scenario was uncommonly tough in that the lens used is the Voigtlander 65mm APO-Lanthar which has extremely low lateral and longitudinal CA, and the out of focus warm coloured autumn leaves on cool green-coloured moss tricks the algorithm even more.
I was just curious as to what scenarios would trigger problems and I would like to be able to identify them better rather than try all the settings. That gets annoying quickly. With so much control, it’s also easy to end up introducing more damage than actually removing any. Knowledge is power…
I cannot agree more. Thank you for turning me on to this. It wasn’t even one of my most important questions, but now I have formed some strong and clear ideas about it. Thank goodness it was kept as an option! Praise to you and the other developers!
It will be interesting to see how it can be tackled. Perhaps a slider to smoothly fade between no colour shift correction and max colour shift correction. The problem is that there isn’t a neutral reference to compare against, and users are (my opinion) getting fooled into perceiving more colour when there’s really none. @TooWaBoo’s example though is a great one for showing when avoiding colour shift is better than not. I have yet to encounter something as severe as that however. But now that I’m paying more attention to it, I’m sure I would find more of such in time.
2 Likes
I can confirm that on images I shot with my Apo-Rodagon 105mm/4.0, which also show almost no CA
1 Like
I will elaborate on the current ‘avoid color shift’ algorithm (which is quite simple) tomorrow. Now it’s too late. Time to sleep…
1 Like
By testing you get the knowledge to use the power of proper dynamic profiles (for example to disable raw ca correction on images shot with an APO lens) 
1 Like
Indeed it is the case, and any helpful hints or tips from people who know better go a long way in helping us all to figure it out more quickly, or even at all. Without your contributions, I likely would not have been as alert to these issues. There are too many different image processing software available, let alone all the various options (and flaws) each one has, it’s totally impossible to know them all. But for the adjustments which affect the fundamental image quality, it’s nice to have a strong grasp on how things work. Have a good rest!
1 Like