Yes, the PDR is essentially the same at the ISOs I’m typically concerned with.
I definitely 100% agree. I edit on a 43 inch screen so any noise is very obvious. DT has great options to reduce the noise. The details threshold slider is my new best friend. I use this for even initial sharpening if the image is noisy. I have just done some 4000 ISO shots from a safari shoot and didn’t feel the need to use the details threshold slider as noise just didn’t seem an issue. However with your playraw bird shot noise would become more apparent in the soft blurred background and this is where you don’t want any sharpening as it will bring out noise but no extra detail. I sometimes use a very rough drawn mask around the subject in this case for both sharpening and denoising.
However, I have not played with AI denoising and can make no valid comments on how good or bad they may be. I limit my comments to DT which now has excellent noise reduction capabilities. When I get home I will look at your posted image.
Canon R7 image shot at 12800 ISO during twilight as a RAW file with no sharpening or denoising.
Processed in DT. Exposure corrected, local contrast applied. Denoised profiled set to 1.5 strength. Demosaicing sharpening AA filter preset (details threshold +94%) and Diffuse or sharpen module denoise preset with details threshold set to -35%.
The presented image is a very small crop at 100% magnification. I come from a film background and I would have been blown away by this result from such high ISO film. The reality is film could not have produced this result. 12800 ISO is not going to be smooth as a baby’s bum, so we have to be realistic.
4 Likes
Hi @Terry, yes that is pretty impressive for ISO 12,800.
As requested, here’s a photo I shot last spring of some Green Winged Teals, shot with my R7 at ISO 4000 and cropped by 50%, which is pretty typical for me. I’ve attached the file if anyone wants to play with it:
619A6302.CR3 (46.0 MB)
This work is licensed under CC BY-SA 4.0 
Here’s the unedited file, exported without and corrective edits from DT:
And here it is with Profiled Denoise applied in multiple selective instances:
619A6302_01.CR3.xmp (34.0 KB)
I like the results, but if I wanted to clarify the background and clean up the residual noise on the duck’s bills and other darker areas, then I’d apply an external NR:
The DT version is probably good for me for most circumstances, but I could do a bit better with some outside help (note that I’m only talking about the noise, and not any exposure or tonal differences between the two, which I could adjust pretty easily).
3 Likes
I will look at this image when I get back home. I have never done multiple instances of Denoise profiled. I am not sure if that is a good idea. However I often use other modules for additional denoising as well as denoise profiled
1 Like
It’ll be waiting for you… 
What I usually do is a single instance to deal with chroma noise globally, then apply another instance to deal with luma noise by masking around the main subject, frequently with NLM. I might even take it a step further by adjusting the mask opacity to “dial in” some additional NR onto the subject. Of course you can do the same using different modules.
I sometimes draw a mask around the subject too for sharpening and denoising. You don’t need to sharpen a blurred background or sky and these are are prone to revealing noise problems. Denoising options in contrast equalizer and the diffuse or sharpen module are pretty good. Luma noise reduction really benefits from the details threshold slider as luma denoising generally works by softening the image.
1 Like
Actually it can be… when DNP went to the new default with chroma and luma in separate tabs this allows for separate color and luminance tweaks in one module but the opacity and strength are the same for both… The chroma noise preset basically has the luminance curve pulled all the way down…
You could do the reverse and create a luma only preset that you could mask separately and use a different slider or opacity setting… I don’t think this presents any issue other than module overhead and in pictures with noise work that needs to be finessed it might be more flexible…
1 Like
Thanks Todd for that information. I will explore the idea of multiple instances of denoised profiled.
Here is a 4000 ISO shot processed in a standard fashion by me and not taking any extra precautions for noise. I will update this post soon with a more carefully edited image to try and minimise noise. This is a very small section of a larger image so I expect some noise to be revealed.
Here is the same edits with the only difference being that I used the details threshold slider to minimise noise by sharpening only the edge details of the subject.
To give some perspective to what I have done here is the whole image. In my opinion this level of noise at 4000 ISO is nothing and doesn’t need AI noise correction. DT handles it really well. I could have done even more to reduce noise, but I didn’t feel the image needed it.
2 Likes
Here is a 32000 ISO image from the Canon R7. I used LMMSE demosaicing for best color. I also used the primary sliders in Sigmoid to further improve color. I used denoised profile at 1.5 strength. I used diffuse or sharpen module and contrast equalizer modules for further denoising.
2R0A0971.CR3.xmp (20.2 KB)
2R0A0971.CR3 (50.6 MB)
This work is licensed under CC BY-SA 4.0
I have been looking for best workflow for same problem. Currently I run Canon Neural Network Image Processing Tool for this kind of images taken with Canon R5 or R8 and export 16 bit TIFF for further processing in darktable. And for images taken with older cameras I run Topaz Photo AI and export DNG for further processing in darktable.
I have found Canon NNIP superior for Canon cameras it supports and it gets even better with supported RF lenses. I have not tested new Lightroom, so can’t compare to that.
Run this image through Canon NNIP, exported TIFF and did some quick edits in darktable and after that some fine tuning with gimp. Would need much more work to get better results. This is just quick test on coffee break.
I feel the Canon software option has produced a superior result. I am unsure how close DT could get to this result. Another subscription of $5 USD would not appeal to many but for others would be a small cost. Your result gives me an aim point to reach for. I actually didn’t apply any sharpening to my image. I feel your result is very relevant to the original OP’s question especially since he has a Canon.
Just for reference, this is what I am getting from DXO and Photoshop
DXO Photolab 6
Photoshop Denoise
1 Like
The moral of the story is don’t shoot at 32000 ISO with an R7 but if you do it becomes a good test case for denoising and sharpening techniques and software. Here is my next attempt with more denoising and some sharpening in DT. Personally I don’t feel it is too bad considering the limits of the camera have been pushed.
2R0A0971.CR3.xmp (35.9 KB)
Pretty awful, looks like a plastic bird 
BTW, I think people are really making the noise too much an issue. I shoot some spectacle at high ISO (>= 6400), yes on screen there is noise. I usually apply the preset on profile denoise for chroma only. And yes there is still noise visible on screen… Now I print in A3 for an exhibition and the print is clean, very good quality, the noise is barely visible and this keeps details which is far more important. It keeps your picture living… So spending between 1 minute and 10 minutes, breaking my RAW workflow to give me a plastic look is not an option to me. Just my 2 cents.
12 Likes
I 100% agree that people get too hung up on noise because screens will let you peek at individual pixels but once you print the image the ink would spread a little bit and hide some noise. Also if you don’t look at an image zoomed at 100% it looks fine often. The original poster has a Canon R7 and was concerned about 2000 ISO. That just isn’t an issue in my opinion. But It did start a nice thread about noise reduction and AI.
I had come to the conclusion that opensource had the denoise covered,more time consuming and to use topaz for upscaling but I might be wrong as I need to do a bit more work on this myself and I’m finding this post helpful.
articles to check
last time I used darktable it was in version 2 . something and unfortunately my subject matter don’t want the image to be public domain though I think it’s good for them myself
I looked in the manual of the time and it suggested using denoise in just one color channel. It worked well
I suspect this might be another case where it is helpful to question the premise of the argument but I wonder if the reason this worked well was that camera sensors work on Bayer filter principles and therefore there is less noise in the green channel. Might it be good to use the complementary colour to green for denoising
I need to do more work on this testing maybe on chroma and luma noise in different channels and color models
also maybe open source / free upscaling is getting there. Best to be on the trailing edge my old friends in open source said
Tried Canon NNIP for this also. Here is 1:1 crop with default settings directly from Canon DPP for you reference.
2 Likes
My sharpening for that image is done way too quickly. It should be easy to improve from what I have done here.
Few years ago I tested Lightroom, DXO, Capture One and some other commercial RAW development software. I liked what DXO did to high iso images but at that time I did have only few high iso images I needed to develop. During my testing darktable got support for Canon CR3 and after a quick test it was clear overall winner for me. Since then have been using DT for my RAW development needs.
I don’t also like at all these subscription services. Unfortunately I could not get good enough results with open source software when I had to develop some high ISO images. So after some more testing decided to give my money to Canon. Before that I had Topaz Denoise AI (some old version) and got few days ago Topaz Photo AI from their black friday sale (99USD for old customers) with one year of updates. Canon NNIP is still a lot better for cameras and lenses it supports (IMO). I have used Photo AI only for some old images to get little better results than I get from darktable. My feelings about Topaz photo AI are bit mixed, it works pretty well for some images, but way too often results are pretty bad and not usable with my standards.