Well, it’s not really correctness; it just means that darktable has valid data from the start of the pipeline. The operations of the pipeline can, of course, ‘distort’ it in various ways (white balance, hue and saturation controls, non-scene-referred modules that clip values, filmic black and white points and their desaturated surroundings etc.). But unless you do something drastic, unclipped raw should end up as something resembling reality. 
I also really need to get my monitors calibrated. I am looking at my post on my PC upstairs and it looks dark and dull compared to what I remember I did last night on the PC in my office… different lighting I guess etc but it sure is a good reminder not to judge anyones edits as it could be the same trying to compare what they saw on the screen when they exported vs how it looks on my screen…
Without reading the full thread…
… I feel this is another case where somebody thinks ‘oh, highlights are blown. Reconstruction in filmic looks like the tool to fix it’.
Highlight Reconstruction in filmic is just that, reconstruction. It is more intended to fix blown out parts with something looking a bit nicer if you have no other choice, as a last resort.
I think - looking from pictures others posted - that your raw contains data there, at least something. Just setting the white slider lower in filmic will ‘access’ that data. Maybe up the latitude to lower the desaturation in the highlights.I don’t think filmic highlight reconstruction is needed here or even does something.
The trick is that filmic might look a bit flat when you want a large part of your highlights in there, which requires fixing local contrast.
As for weird colors in the highlights, try messing with different settings of the highlight recovery module to see if one of them is better, and also try it off.
1 Like
that’s exactly what I thought!
thanks for the tips, I will have another look at my process!
Didn’t go ‘creative’, just an example of what ‘exposure + filmic, + local contrast + color-balance’ can do.
dt 3.7.0+1090
20120116_D7K0879_1.tif (12.2 MB)
I did do my ‘cheat’ again. I loaded the raw file into DxO Photolab to get optical corrections. In this case I did a tiny bit of straightening (took a window from the building right in the center, and wanted it to go straight up), and added ‘auto’ perspective correction which did a very, very tiny amount somewhere.
Other DxO auto tools are chroma-denoising (luma I turned off) with it’s old-skool mode (ISO 125, no need to go fancy here), vignetting and lens distortion, lens-sharpening and maybe a little cheat: ‘micro contrast’ at it’s default detection.
Exposure was off, white-balance was off, tonal adjustments were off, color rendering was off. So you get as much of the pure RAW in the demosaiced DNG. It’s just ‘fixing things’, all the tonal changes I do in Filmulator or in this case, Darktable.
20120116_D7K0879.dng.xmp (17.4 KB)
20120116_D7K0879.dng (60.0 MB)
Darktable loads per default with filmic off and highlight construction on, but the color-workflow is modern (so white balance to ‘reference’ and ‘color calibration’ is enabled with the stock camera-detection. I do turn off the ‘highlight reconstruction’ module because I do not think it’s needed (yet). It looks like you did a good job for exposing for the highlights.
With filmic still off, I increase exposure till I think it sits somewhere nice. I enable filmic, raise the lattitude and then hit the ‘auto’. It works (yayfor denoising) but I still lower (‘to the right’ at least) the white slider in filmic till I get color back in the sun. The only thing clipped in your RAW was the very small sun-disc. The surrounding parts are only clipped in one channel, or are ‘almost’ clipped.
It now looks very flat, so I try the tone-equalizer with the preset ‘contrast curve medium’. I move the 0ev point back to 0, and I do think the shadows where a bit too dark so I make the ‘darkening’ parts of the curve a bit less.
Still needs a bit of local contrast for my taste, so I grab contrast-equalizer. I do not want to touch the very fine parts there, because then it becomes more of a sharpening effect. I also don’t want to push the very coarse side because I’m looking for the details in between the very global contrast and the micro contrast. So I push the middle parts up a bunch:
It can look quite overdone, that’s where the ‘mix’ slider is for underneath the curve to lower the effect to your liking.
As a bit of advanced thing: I still felt it crushes the shadows too much, so I enabled the parametric-mask and pushed the sliders up on the ‘L’ section (The L from LAB, so luminance) to reduce the effect in the very dark parts.
I still like more details / clarity in the highlights, so I do my bilateral trick there. Enable local contrast, bilateral, very low contrast, very high detail, use parametric mask to only select the clouds. Tweak the ‘blend mode opacity’ to taste after to lessen the effect. In this case, I left it at 1.0.
In the end (maybe it was sooner, I don’t know) I enable ‘color balance rgb’ with the ‘add basic colorfulness’ preset and start moving things where I think they need it on the master tab. In this case, I felt like I still wanted more color, so I increased the global vibrance and global saturation. I like the effect of increasing the global brilliance (or only in the mid + highlights) a bit, so I did that as well. I still felt it wasn’t doing that much… maybe just not a lot of color in the image. Anyway, I just left it there.
Nothing fancy, just a way to show that if you set exposure first, then filmic to ‘pull in highlights’, then local contrast to taste you’ll get a lot of the way there without (much) fighting. Creative edits are up to you, of course.
4 Likes
wow thanks joris, so much information here! I love when somebody details his approach – lots to learn!
I discovered the color reconstruction module taking a look at your edit and i find easier to manage to recover the highlight as the filmic rgb reconstruction tab. Thanks! 
Using dt 4.1 with the new segmentation based highlight reconstruction. It worked great, although I need to learn more how it should be used - it’s mostly trial and error atm 
20120116_D7K0879.nef.xmp (28.2 KB)
1 Like
About trial & error😎
-
The combining should only be used if you have blown-out parts of the image - let’s say two segments A and B - very close to each other and either A or B is surrounded by a pretty dark border. Think of a mesh for example.
-
Candidating is to be tried. A general rule, if you have smooth transitions from bright to blown out like in the sky or clowds you are fine with Candidating like 0.4, if mostly have harsh transitions like reflections or a sight through a window where outside is blown out, candidation should be very low, close to or even zero.
4 Likes
Got it! Thanks very much for the tips. I’ve forgotten which slider it was now - but one I had to lower to near zero before I got rid of pink bits around the sun, where (I think) all three channels were clipped. I suppose that would be combine…
(I’m on my laptop with only 4.0 atm so can’t check)
I had another play with this file in a modern Darktable. I see my last attempt was somewhere between 3.6 and 3.8, feels like some time ago .
I disable as much as possible and lower the exposure till all the highlights fit. I see this:
That magenta in there seems to tell me there are clipped highlights in this file, so I was wrong before.
If I enable the raw-clipping indicator:
… still a bit of the magenta is not marked as ‘clipped’?
So maybe the white level is wrong in this file, like the recurring theme it is…
Seeing no real info about white-level with exiftool, I check the camconst.json file from Rawtherapee:
Darktable uses 15892 for it’s white point. Looking at RT’s camconst.json, the correct answer for this camera seems to be “it is complicated”…
I start lowering the white-point bit by bit with the scroll wheel of my mouse, but with the raw clipping indicator still enabled. At a white level of 15836, the clipping indicator shows something that makes more sense:
Reconstruction is needed:
- clipped is ugly
- ‘reconstruct lch’ removes the color
- ‘reconstruct color’ gives artifacts in this file.
- laplacians seem very promising with the default 128px and the iterations to 16… but the large glow isn’t really ‘reconstructed’. Increasing the radius turns the small sun disc into something uglier and makes it slower… In the end, couldn’t get a result out of it that I liked.
So… new segmentation it is. The defaults already do something quite nice, but leave the sun disc as magenta. Messing with the ‘candidating’ slider gives me a nice reconstruction at around 4%:
So, now the highlights are fixed. We can properly ‘start’ with the file 
.
- Exposure to my ‘default’ of 0.7
- Filmic with my ‘Aurélien’ defaults: contrast 1.0, latitude 0.1, v6, maxrgb mode, hard contrast in shadows + highlights.
- I hit the ‘auto white picker’, leaving the blacks alone.
I end up with something like this:
Not bad for a few clicks.
- Add ‘color balance rgb’ with the vibrant preset, turn the highlight desaturation back to 0% on it (seems nice around the sun and it can take it).
- Add default ‘local contast’
- Add ‘local contast bilateral’ masked to the highlights to add some drama to the clouds. I make sure the mask doesn’t extend to the reconstructed highlights, because it can show some weirdness there.
- Enable lens correction + raw chromatic aberrations (really needed since the lens correction doesn’t do TCA it seems!)
- Add a simple tone-equalizer to lift the shadows a bit.
- Add diffuse (x2) for sharpening.
- At the end - as an experiment - I use the ‘clarity’ preset on the ‘contrast equalizer’ to get some more local contrast. I enable the mask to not let it work on the highlights (in the ‘g’ channel mode), because it again messes with the gradients in the reconstructed highlights (maybe it even clips the extreme highlights again? It’s a scene-referred but LAB module, after all…).
- And I added a ‘subtle saturation + warming’ LUT.
Now… in the end, I reset the white-point back to default to see what difference it makes… none. So I guess the whole white-point intro wasn’t needed .
Still a nice result (IMHO) without my ‘DxO cheating’ I did a year ago.
Now, I normally export in linear-rec2020, do a downscale, and convert it with imagemagick back to sRGBv4 with relative intent.
But the export (it seems already from Darktable) is crazy saturated in the reconstructed highlights if I do that. Like really yellow. No artifacts, but just really yellow. Not at all what I see in Darktable.
What is going on here?! (export left, Darktable window right).
I settled on exporting ‘linear 709’ and just interpreting it as ‘linear rgb’ (sRGB-elle-V4-g10.icc).
20120116_D7K0879_01.nef.xmp (78.8 KB)
edit (reupload): The aberrations in this file make me go nuts. It seems to be localized on the left side of the image? The edges of the towers on the left are full of yellow / blue. When I use ‘TCA override’ in the lens correction, I can get rid of it, but the noise in the sky above it goes crazy.
If I enable ‘raw chromatic aberrations’ it seems to clean it up nicely… but then when looking the sky on the right, it added aberrations all over the clouds.
If I use the (non-raw) ‘chromatic aberrations’ module, it seems to get rid of it, but also removes a large portion of color in the small details all around. I ended up masking a bunch (in a crude way :P) to let it only work on the biggest offenders. But argh. Is this a lens issue? The fact that the lens profile has no TCA data maybe also means’ this cannot be autocorrected’ ?
6 Likes
An older thread with a lovely shot. Not seeing a B&W offering, I thought I’d offer up a contribution…
Edits via ART:
20120116_D7K0879.nef.arp (36.8 KB)
My objective was to bring out greater contrast in both the sky and ground surfaces while also emphasizing the sun’s reflections on the tops of the buildings to the left of the scene.
ART’s local contrast and tone tools make selective masking readily accessible for such editing.
Thanks for the chance to enjoy this handsome shot taken years ago.