DT Noise reduction poor results – Advice requested

I’m following Bruce William’s excellent YouTube series on DarkTable in a (very slow) attempt to learn this highly functional product. I have tried to reproduce the results he obtained in Episode 19 (Noise Reduction) particularly when processing an extreme example of colour and luminance noise. I am not getting acceptable results, so would appreciate some advice. I am conscious that he used DT 2.6; I’m using DT 3.2.1, but they’re both Linux versions, under Mint.

Bruce stated that, by comparison, LightRoom was ‘not even close’ to what he achieved using DT on an extreme example. I quite share that view – as long as I interchange the words ‘LightRoom’ and ‘DT’, so clearly I am doing something wrong because there is such a huge difference in our results.

I have attempted to include crops from the best versions of images I was able to obtain, exported (as jpegs) from Lightroom and DarkTable. The DarkTable version uses only Denoise (Profiled), but I had tried all the other (total 6) modules suggested by Bruce, singly and in combination. The Denoise (Profiled) option appeared to be the best I could do, but even then it look as if somebody has gone berserk in this church building with a 3-colour spray gun! In case I haven’t clearly labelled the crops, the one with the obvious noise is the DT version.

Any suggestions about other things I should be checking, such as other modules in the stack, or the use of CL?

Perhaps worth checking out some of @rawfiner’s videos. For example: https://www.youtube.com/watch?v=7ZhbeXpx2W8

While the lightroom version does look less noisy, you’ve also lost a lot of detail and it looks quite waxy. Perhaps you could post a PlayRaw and see what the community can achieve.

2 Likes

If you really want to compare the two programs, please make sure that exposure and white balance are as similar as possible between the two images: in this case, the DT version is clearly warmer and has a higher exposure (==more amplification of noise).

Also, it’s hard to suggest where you can improve the DT treatment, as you don’t give any details. A screenshot of the settings you use in the denoise module (and perhaps the history stack) would be very useful.
Alternatively, you could post the raw + the xmp file from dt (with a proper license statement, see the “playraw” section here), so others here can have a go.

1 Like

Please share a this raw file so we can help you better.
We can’t do much without a raw file.
I promise you, DT has better noise reduction than LT. You just have to know how to achieve the desired result.

3 Likes

Thank for these comments. Yes, the LR result is not perfect; far from it. It was just a quick attempt (about 2 minutes of work) to see what I could achieve quickly on this old image - a sort of basis for comparison.

I would be most grateful if community member could work on the original, so I will try to understand what ‘PlayRaw’ means, as it’s a term that is currently unknown to me. Thank you for the pointer to other videos.

There is a subsection of this forum, Processing > PlayRaw. Make a new post, upload your raw files, assign a license to the raw files, and write a bit about what you want to achieve, then people will edit that file and provide you the sidecar, which you can download and load into your editor to inspect what they’ve done.

You are quite right: I haven’t given enough detail - mainly because, without asking first, I really did not understand what detail would be helpful/necessary for better informed community members to make a meaningful comment. Now I’ve learned a little and will post some screen shots. Yes, I’m going to try to post the raw xmp files (the DT one, not the LR one) to the community once I have understood how to do this.

See also PlayRaw stuff to keep in mind

Yes, having seen what Bruce Williams was able to achieve, I am sure DT can do a better job than LR - right now I’m not sure I can ‘ask’ DT, in the right way, to do it! I’ll be posting a raw file just as soon as I’m confident of being able to do it.

I have taken 2 screen shots, the earlier one showing the contents of the history stack (plus the image) and the later one showing the settings I applied in Denoise (Profiled). I have failed to understand how to license these images so I cannot use PlayRaw yet. In the meantime, I would say there are two major objectives for the post processing of this mage that I would like to achieve: 1). Reduce the luminance and chrominance noise; 2). Recover the blown out detail in the windows - these have intricate metal tracery in them which I was able to reveal very quickly with one single adjustment to the highlights in LR.

In your play raw post say “images licensed under <license name, preferably some CC license>” and that is it.

@LateJunction, you probably haven’t reached to that episode yet, but earlier this year Bruce Williams revisited the “denoise (profiled)” module with dt 3.0:

Some of the advice from version 2.6 has changed since that time.

Just a quick advice from what I find in your screenshot of the history.

you are still using old modules like basecurve and sharpen, which are part of display-referred workflow in Lab space. with dt 3.2.1 you probably should/want switch to the new scene-referred workflow (Exposure + Filmic) using a linear RGB pipeline. That will improve a lot of your possibilities to recover highlights (2nd objective).

For denoise (profiled) I prefer nonlocalmeans mode and a single instance. a good starting point is the auto preset. set autoadjustment slider to same value as your exposure compensation for this picture (compensate visually for middle grey in combination with Filmic). Many times this may be sufficient. if not: switch to manual mode and tweak the result.

you should watch the explanation vids from the devs

2 Likes

denoise (profiled) in non local means mode (or non local means auto) with setting central pixel weight (details) to 1.0 might be an option to suppress the chroma noise

2 Likes

In this screenshot:

  • you should keep the U0V0 curve normal or increase it as it affects chrominance noise reduction
  • you can lower the Y0 curve if you want to preverve luminance noise
  • no need to use a color blending mode. The Y0U0V0 mode already allows to reduce separately chrominance noise from luminance noise
  • you can increase strength (even values higher than 4 are possible using right-click with a keyboard entry)
2 Likes

Thanks, I’ll try that shortly.

I think I was deterred from trying to use scene-referred workflow when I found that “exposure plus filmic” produced a result on another test image (portrait shot of two young friends) which was so extremely ‘plastic’ looking, where as a quick adjustment in LR produced a result which was very satisfying indeed. But I hear what you say and will go back to have another attempt at this, reflecting the other advice I have received. All of it most helpful.

Many thanks for this - you are right, I had not reached that episode yet but now I’ll look at it and try again.

OK, I botched together a new post in the ‘Processing>Play Raw’ category, copying somebody else’s license statement and included links to the raw file and the (DT) xmp. But, embarrassingly, I was unable to figure out how to include a view of the image in the post. :confused:

Just drag-and-drop the exported jpeg into the post.