Experimental: you could try inpainting the dark spots; e.g., heal tool, G’MIC, …
@afre Couldn’t find the setting to make it public 
The only thing I explicitly made secret was my e-mail (I read “Never shown to the public.”)
Hum, right, but for now I’m lazy 
Thanks for the idea anyway
Could you send the xmp for above file pls!
To mee it looks like, I had to boost exposure by 1 and still my histogram looks diff than yours, also strange, your preview shows 100% but this seems to be a 200% zoom
+1EV means, we are at ISO6400. Unfortunately for your camera, there is no such denoise profile. Acc. to Harry Durgin, it helps to change in that “found match for…” on top left of denoise_profiled.
You used channel mixer to B/W, so do I and I fiddled around with the different options. On purpose I did not compress my history stack. To me "soft skin ever looks not nice, hence I used Kodak T-Max…
I tried the monochrome module and it produces a bit splotchier result.
This is, what I achieved and the efforts is similar for what I do with my pics. Around step 15 or so, I would have stopped… Later I will check, what my D750 says…
_MG_5113.cr2.xmp (12,5 KB)
what screen resolution do you run at???
@gadolf And that might be one reason… hence I am looking for your xmp of the file you provided…
symptoms fighting? Also way too soft in my eyes 
These dark blotches are, I think, due to photon noise (Shot noise - Wikipedia)
Basically, few photons hit that pixel, so the pixel is completely black.
If the pixel is black only on one of the 3 channels, the denoiseprofile can lead to ok results.
But denoiseprofile is generally struggling at lot at removing pixels that are black on all 3 channels.
My advice here is to use denoise bilateral, which smooths more dark areas than bright ones. It can be used combined with denoiseprofile or alone.
If you are using darktable dev build, you can also try putting an instance of denoise bilateral before denoise profile, it works quite well.
Also if you are using darktable dev build, for denoiseprofile, you can try using non local means with a big patch size (the parameter value can take values in [0,10] with right clic), setting the scattering parameter so that few chroma noise remain visible, and setting the details parameter at the end to get back some details 
Hope this will help 
2 Likes
now I see…
- you boosted exposure by 2EV that equals to ISO12800!!!
- go to channel mixer, set it to grey and reduce red from 1 down to .5 or so and lift green and blue ~.25 or do all 1/3. You will see a much better result!
In fact!!
The point is that I don’t want to do a monochrome edit. The images I sent were both monochrome because I was in the middle of @rawfiner’s tutorial, where you temporarily switch to monochrome to be able to denoise each rgb channel separately (as previously mentioned by @afre, here)
I’ll keep experimenting with all suggestions and will later state my findings.
EDIT: @AxelG given that I want a color edit, I’m stuck at you advice to tweak the channel mixer channels, regardless the fact that it does what I want in monochrome.
This part was not clear to me…
I thought you where aiming B/W
Than I have to see again in color…
I am in a hurry now, so sorry for quickies…
Try this xmp
- zoom as you did
- switch off chromatic aberations, hot pixles and 3 instance of denoise-profile (denoise_2 is with L-mask)
- switch amaze back to ppg
- do a snapshot
- switch all those corrections on again (and play with ppg/amaze)
- move the snapshot slider and you will see…
it is in your sensor data (indeed) and has a very hard saturation (figured while I wanted to mask it), even at 99 of upper right input slider of “S” those pixls which makes your splotches, are out of the mask, right away)
Usually amaze (implemented here with two times smoothing) helps a lot but also costs colour…
try to borrow another cam and shot the same scene with both cams ( or three, one as yours and one different one, e.g. the Nikon D750, which is an ISO monster)
_MG_5113.cr2.xmp (32,9 KB)
The more I look, I feel, it is your sensor, sorry to say
that is a 400% zoom. on the left
_MG_5113_02.cr2.xmp (53,9 KB)
on the right, I switched off color channel and all denoising and than move the snapshot slider
you see, that those spots, black on the left, they are also black on the right (I know, above I talked about over saturated. This part I didn’t bring together yet, as I gotta go now…
I find you could get somewhat get better results when increasing brightness by using a tone curve instaed of the exposure module. The crop below shows the red channel using the biliteral denoise module.
_MG_5113.cr2.xmp (2.2 KB)
It did!
Before:
After:
I loose some detail on the arm’s hair though (best noticed while zooming out), but I guess it’s a matter of further tweaking to find the best trade-off - or even masking it out.
Before:
After:
I also used luminosity masking while applying denoise bilateral to preserve more details:
From bottom to top:
First denoise bilateral instance,
Second instance:
With this, I tried to mitigate the plastic look produced by this module.
Follows xmp file: _MG_5113_03.cr2.xmp (23.2 KB)
Thanks for the tip!
Interesting.
I moved the exposure module up the pipe (after denoise bilateral), and it gives similar results to yours, regarding noise:
I’m a bit cautious to switch exposure for a tone curve because of this
In summary, I’m confused with your approach.
EDIT: Please correct me if I’m wrong, but my understanding is that it is important to set up the right exposure at the beginning of the pixel pipe because the following modules will profit from a broad histogram (comes to mind the filmic module). But it is also important that this is done in a linear way, so that tone mapping modules can start from scratch. Is this layman’s interpretation correct?
Applied a box filter with edge awareness for demonstration, so of course it won’t be as nice as the bilateral filter as @rawfiner suggested later. Even then, should I tweak its parameters, I am sure that I could come close to something satisfactory.
The curve gives you an opportunity to decide which tones to lift. As for linear v nonlinear, and clipping and saturation, in a raw processor, the devs control the pipe, so although you have some choices, they aren’t many. I don’t typically use dt, so I cannot provide dt examples or give you advice on dt processing.
So you’re saying that if the pipe is fixed - which I believe it isn’tsince DT 2.7 - it doesn’t matter if you start using the top most modules and proceed down below? I got the impression that you should go from bottom to top as a general rule. (ok, sometimes I change the settings of a bottom module after activating many modules above, but I do that with care)
The order might be adjustable but
Devs are always wrestling with this tension. In GIMP e.g., you see options starting to pop up for the user to adjust how the tools behave.
This is one of the main reasons for my hack software, rawproc - I wanted to insert the tools in arbitrary order. I also wanted to be able to omit tools. If G’MIC would have opened raw files without a pipe from dcraw, I might not have started developing rawproc.
I should have read this before… way more competent than what I tried to express, even though I was close 
