Highlights recovery


I understand your point and I tend to agree. For what concern the highlights, though, I think is not crucial.
Trying to match everything makes sense when it comes to color rendition. I can try to adjust ACR as much as I can but I’m pretty confident the result won’t change much in terms of how much I can push the highlights.
Anyway, I can do a test and I’ll post it later.


Highlight recovery is a great topic to explore. I encourage you to participate in the PlayRaws. I learned most of what I know by developing those raw files. :wink: Plus, many of the participants have kindly provided their sidecar .xmp for you to examine.


Thanks, I’ll check the PlayRaws.

In the meantime, I did another test with ACR, setting everything to as neutral as possible.
Here’s the before:

And this is the after:

I deliberately pushed the highlights to the extreme.
As I thought, neutral or not doesn’t make much difference in terms of highlight recovery. I don’t know the technicality but from my experience, the capacity of a raw development software to recover the highlights hasn’t much to do with the profile. What the profile can do, though, is providing the user a more neutral starting point hence the user has to deal with possibly a more balanced and less extreme highlight.
In other words, it can be useful to set to neutral DT rather then ACR and see if at that point, instead of dealing with blown out highlights, it’s just a matter of bringing back some contrast and, the hardest part, reconstructing the profile. Hope that makes sense :slight_smile:


My standard approach: lowering exposure + highlight reconstruction in LCh.
Switched on Purple fringing and demosaicing against colour artifacts. Tone curve lo lift up the shadows and punch up colours.

_DSF0498.RAF.xmp (5.6 KB)

(Gustavo Adolfo) #25

This seems to be very close to what ACR did (as shown in your last post), isn’t it?

_DSF0498.RAF.xmp (7.1 KB)


Thank you, Jürgen

I’ll take a look at your xmp file.
Thanks for sharing it!!


Thanks, Gustavo,

That looks close to the ACR version.
I have looked at your xmp file and it’s been helpful to see how you deal with the overexposed area, a good combo of the shadows/highlights & the exposure modules.
One thing caught my attention with this technique is that the image loose details quite soon.

Maybe it’s a matter of tinkering with the image. I’ll keep testing, it’s definitely interesting to see what is possible to achieve.

Thank you all for your help and advice!

(Gustavo Adolfo) #28

Can you upload the full res ACR jpg?

(Peter) #29

Jürgen - Impressive!
Two questions:

  1. you don’t have any “default” settings: orientation, sharpen, base curve. Did you removed them?
  2. first thing you have in xmp workflow is raw black/white point. You added this or is it default when you open RAF? If default, then why I don’t see it in my RW2 files? Is it applied automatically or should I remember about it and put in my workflow?
  3. you change tone curve, and then still pull up shadows in “shadows and highlights”. Why this method?
    General questions:
  4. in demosaic there is no more PPG / AMAZE. VNG4 is now VNG. Removed because of poor results or missing in 2.4.4?
  5. is DT 2.5.0 available for Windows, and if its ready where can I download it?


I posted too quickly…
Yes, I’ll upload the jpg from the ACR tonight when I get home. Thanks!


Hi Gobo,
1 => Yes, I started from scratch / original.
2 => raw black/white point: I activated this one to see if there is some “headroom” left in whitepoint adjustment. No changes,can be omitted.
3 => lifting up the shadows in “shadows an highlights” is different from applying the tonecurve and close to the same function in lightroom. drawback: prone to produce halos. I use it only in low amounts < 40 with all other sliders set to neutral.
4 => this is because of the x-trans sensor of the camera. If you develop a RAW from e.g. CANON, NIKON the other options will be available.
5 => I think 2.4.4 is still the actual release.

(Denis Testemale) #33

Sorry for not being able to propose my own version but I don’t have access to my computer now. In the case of difficult highlight recovery I find that removing the basecurve does help a lot : you then have to work harder on the midtones and blacks but at least the highlights are untouched. Maybe you could try the multiple exposures blending from the basecurve options.
Again sorry I can’t check on the raw file, but if I understood correctly Jurgen did start from scratch as well, i.e. without the basecurve.



There isn’t much detail to begin with in that region. Mosaic at 500%


@afre That’s not the part I’m referring to.
In the image I posted above I circled the side of the building. Initially there were lines, they are now basically gone.


Thanks for your input, Denis.
Yes, that was my point as well when I wrote this:
“In other words, it can be useful to set to neutral DT rather then ACR and see if at that point, instead of dealing with blown out highlights, it’s just a matter of bringing back some contrast and, the hardest part, reconstructing the profile.”

(Glenn Butcher) #37

Thought I’d follow @afre’s lead, drill down to the rawest data. This screenshot shows your raw file opened with libraw’s rawdata option; this is the image array pulled right out of the RAF, with absolutely no processing:

Note the histogram; there’s a lot of data piled up at 64 (histogram is scaled to 256 for convenience), 64 * 256 = 16384, which is the saturation limit of your 14-bit sensor. What can’t be known now is how far past saturation those pixels were in the real scene, making them of no value to anything purporting to ‘reconstruct’.

It’s not easy to convince a raw processor to show you this. Most want to give you something presentable as a starting point, which usually involves white balance, demosaic, and some sort of scaling to fill the display range. Even then, finding it out when you get home and start processing is a little late to do something about it ‘in the scene’. And, in a scene such as this, exposing to capture the sunlit data is going to put the challenges on the other end, in teasing details out of the shadows at the expense of noise.

For what it’s worth…


Yes. While the wall on the right has its joints smudged, the wall in the sun, the highlights part of the image, doesn’t have much texture remaining in the brightest sections. Whatever recovery you do would be the propagation kind where you bleed colours from adjacent regions. At this point, it might be better to clone or re-synthesize the textures in GIMP.

(Gustavo Adolfo) #39

I did some more tinkering and could retrieve more detail from those sections you’ve marked, while still keeping resemblance to ACR result:

_DSF0498.RAF.xmp (15.9 KB)

If you could upload the full res jpg from ACR it would be great, to see what level of detail is being revealed by ACR (or being lost by DT). In the last image you’ve posted it’s hard to see any detail.

Inside the red circles.


I am not as familiar with darktable as other people. I used PhotoFlow and G’MIC instead. I kept the processing to a minimum. The relevant part: I applied two curves, one darkening the image and the other brightening a copy. Then, I averaged the two. It is possible to do it with a single curve but it is easier for me to use two, each targeting a different focal point.


Thanks again, Gustavo

I have the jpg ready. It’s nothing fancy in terms of CC, just some quick tweaking of the default ACR settings but I think it should be good enough for you to play with DT.

In the meanwhile, I wanted to thank you all people for helping me out with this. I truly appreciate every hint and advice from all you guys. I know it takes personal time and effort and nowadays sometimes help might be taken for granted. So, just wanted to say thanks :smiley: