extremely clipping image: ART or darktabe or RawTherapee? [solved]


Ks.RW2.xmp (9.6 KB)


Ks-1.jpg.out.arp (11.6 KB)

1 Like

Hello apostel,
very interesting to see: Apparently you can achieve identical results with both raw developers.
Let me know what your experience is. Are they both equivalent at extreme exposures? Is it just a matter of taste which one you prefer to develop with?


What were your main steps?
What did you use to reduce clipping?

You can follow my steps in the .xmp file :slight_smile:
Perhaps I should have included whitebalance as well…

I think that with both apps I can achieve results of equal quality ( judging by my taste) in pretty much every situation. I think ART is easier since there are more automatisms (or “hardcoded stuff”) under the hood that just work well for me. In DT you have to take care of more stuff. Although there are news to the hl-rec-module that clearly help handling highlights (and their magentas).

1 Like

Yes, I am following your xmp file and am amazed. Your approach makes me think. How simple and how good at the same time.
You seem to have done little with “filmic rgb” and nothing with the great “color balance rgb”, and accomplished a great deal.

Whitebalance - that’s another topic, we can leave it out here, anyone can easily improve that.
But what is totally interesting, and I ask you to try it out with you: Select “reconstruct in LCh” in “highlight reconstruction”, the picture will be much better. Try it!

1 Like

A good opportunity to test darktable master with sigmoid module.

Ks.RW2.xmp (7.8 KB)

1 Like

The old HLR methods are not really suggested… if using filmic and modern wb in DT the assumption for their use is no longer really technically correct… AP explains this in his GLP video… so far I find the new method much better…

what is it: sigmoid module?

I would have to open both sidecars but you don’t look like you exposure matched them which would be a bit important to compare them…the DT image is lifted far more inside which seems to be translated to the outside… just my quick observation…

there is so much info, it sounds very interesting, but it overwhelms me

What is the old HLR mehtods?

What video?

I tried the new method and it is not so easy to get good results. Can you show me how to do this?

darktable 4.0.1

Ks_01.RW2.xmp (14.7 KB)

Hello @mperok
Good result. Why did you use tone equalizer twice? Can’t you create the values with one instance?

No filmic or sigmoid… :slight_smile:

Ks.RW2.xmp (11.3 KB)

1 Like

One common way to edit can be to use one tone eq instance to compress the shadows and highlights and then a second to add it back… I have often used 4 or 5… esp when blending in a single color channel…say blue to tweak a sky…

Also a good result with a mysterious light on the vegetables in the green boxes.
What’s the point of using the tone equalizer twice if you pick up something from the first again with the second instance?

And also with your version the result can be much better if you use in "highlight reconstruction: reconscruction color. Try it out!!
And, why do you omit the great “fimic rgb”? Don’t you like it?

It is very interesting that obviously all roads lead to Rome.

I lost way too much detail on the walk way with the tone mappers and they arent necessary… as for light my monitor is sadly not calibrated …not sure about yours so subtle tones and color casts might be very different for us…

EDIT : Think of the serial processing of data…you flatten with one instance then with the new data you target the image for contrast…you will not always be able to get something like this in one instanse of the tone eq infact it is one of the biggest problems I think from users is they try to get too much from one instance and end up killing the image…

I think this refers to the highlight reconstruction methods “clip highlights”, “reconstruct in LCh”, and “reconstruct color” in the respective module. These methods assume that highlight reconstruction will be applied after white balance, which is no longer true for the modern workflow with the “color calibration” module performing the actual white balance, and the “white balance” module being more or less bypassed.

There is a new method in 4.x, “guided laplacians,” which can work well for little blown out areas, but is computationally very demanding for real world cases (e.g. 25 iterations, diameter 1024 pixels).

A less demanding new method will likely be introduced in the next release IIRC. For now, to circumvent the issues, there is the reconstruction tab in filmic rgb which can sometimes also help with highlight issues, but all of this requires some work, as parameters have to be set.

I would guess it is one of these on Aurelien Pierres youtube channel, these are quite good but also quite technical and full of rants.

As I said, this requires some trial-and-error. What I do is checking if I can solve the issue with filmic reconstruction alone, with guided laplacians alone, with both of them, or if I have to ditch the highlight detail entirely and go back to legacy “clip highlights” without “color calibration.” In some rare cases, even switching all highlight reconstruction off helps.

1 Like

Hello, @chris
Your words are very reassuring.
The problem I faced is this: There is something very fascinating about ART because you get good results so quickly. But I decided to switch from RT to dt because of the many possibilities and because I assume that dt will be well developed.
But in an overexposed image I saw that ART very quickly gave a better result than I got with dt. This made me very suspicious. But now I see from the examples that you can get the same good images with both programs.

I understood your words well, except for your ideas in the last paragraph.

There are inpaint opposed and segmentation based already in the dev builds. Both should do well on this photo. I used inpaint opposed on my edit.

Oh my! Now I quickly developed this photo with my old RawTherapee 5.8. In a few seconds I had a result that leaves all attempts with ART and dt clearly behind.
I pray that RT wants to stay alive after all and I will seriously reconsider my intention to leave RT.
Ks.RW2.pp3 (12.9 KB)

1 Like