2 bugs I've found in ART

  1. Open this photo play with sunset colours and use Spot White Balance on around x: 1000 and y: 3000, then look at the clouds around x: 2360 and y: 1560. If the Highlight reconstruction is set to “Off” or “Color Propagation” everything looks fine, however if it is set to “Blend” unnatural orange color appears

  2. Flat-Field embedded in metadata affects highlight reconstruction. For example, open this image https://img.photographyblog.com/reviews/google_pixel_5/sample_images/google_pixel_5_22.dng and apply the Neutral profile, then zoom in to the sun. Not only we see that Blend differs from Off and Color propagation like in bug 1, but also if we enable Flat-Field embedded in metadata we get more of that unnatural orange if the Highlight reconstruction is set to Blend. Even if we switch the Highlight reconstruction to Color Propagation and lower the exposure, we will see that turning on Flat-Field adds some kind of pixel artifacts

ART was compiled from the latest source on BitBucket

and thanks for the feedback!

highlight reconstruction, by its nature, always involves a bit of guesswork. That’s why there are multiple choices available, as there is no silver bullet (at least, there is none in ART’s bag of tricks). I wouldn’t consider these issues as “bugs”, in the sense that I’m not seeing any evidence that the tool is misbehaving. It’s just that different algorithms produce different results.

Regarding flat-field correction, the fact that it might affect highlight recovery is unfortunately an unwanted side effect of how this is implemented. I agree it’s not optimal, but it’s not something that can be solved easily without a major refactoring I’m afraid.



Thank you

Could you please add the “Clip” highlight reconstruction to ART from RawTherapee for cases like these?

As you made reference to this thread,
As highlight reconstruction is not hard science but rather a guessing approximation, you will be able for each method to find some photos where it fails. I should not say it is a bug but rather that you cannot apply the considered method on this specific photo.
So, apply the best method (as you judge it) to the photo when needed.

1- I observe that highlight reconstruction in ART is equivalent to the corresponding one in RT using “Clip out of gamut colors”.
2- regarding flat field correction with embedded flat field, this doesn’t exit in RT, so you cannot observe the influence of Clip out of gamut colors on this corrected image.

Conclusion, use the algorithm which provides the best result.