Denoise Comparison - Darktable, ON1 and Topaz

Even so your edit all in all is nice, it shows well a problem that darktable’s chroma noise reduction quite often produces: Somebody has hit your duck on it’s head. some bits of the head got replaced by the water of the background (unfortunately I often struggle with this problem as well):

Screenshot_20231125_172658

I think that problem was not visible unless one goes pixel-peeping.

One can tweak the detail params, and/or use more careful masking.
Tweaking the params (reduces noise suppression):

More careful masking:
image

And an alternative version (see sidecar):

(Ignore the blurred area on right-hand-side: darktable now uses a low-res preview if the image gets moved, and I did not wait for the full rendering to finish.)

0P9A0410_01.CR2.xmp (17.9 KB)

1 Like

You are right, that one can see this problem on this picture only if you zoom in. And you are right, that there are ways to work around. Anyway, the ways to work around are often time intensive and the problem often occurs to a bigger extend. I love the possibilities darktable offers on noise reduction. Anyway I’m sometimes annoyed, that I have to spend time on masking out such areas to avoid this behaviour. Usually I want to be fast with raw editing.

Hello everyone,

Here is the result from DXO Photolab (newest version, 7.1) with its algorithm (Deep Prime) to denoise RAW images.
Just used the default settings: opened the Raw and denoised.

I have exported this Raw to Jpeg to save some space on this forum (the quality is worse than Tiff…).
On my computer (powerful Intel CPU and 32 gb of Ram) this RAw only takes a few seconds to export.

2 Likes

That is very nice. Thanks for sharing

Very interesting thread. I have found “Ian’s 2019 noise reduction” in G’MIC to give quite good results although I don’t really understand what some of the sliders do. It’s also slow compared to denoise profiled in Darktable.

1 Like

Just for comparison’s sake here is Neat Image V9. I used LMMSE demosaic and disabled sigmoid before exporting as a linear tiff. Then used Neat Image, which you can select an area to profile. But I let it auto select it, which provided a good sample I thought. I made some minor tweaks increasing luminescence, cr/cb chrominance noise reduction. I turned on its default sharpening and detail recovery. I used DenoiseAI-low-light side car adjusting only the WB.

2 Likes


0P9A0410.CR2.xmp (12.9 KB)

DT 4.4.2

2 Likes

I have to agree. I get bombarded with Topaz ads on youtube and I was shocked to see how much better dt is than Topaz.
Thanks for the comparison

3 Likes

Sorry, I’ve just realised the masks were completely ridiculous in this example. :frowning:

Testing Noise Reduction in ART 1.21 on macOS:


0P9A0410-2.jpg.out.arp (11.2 KB)

PS: White Balance taken on grey part of underwing.

1 Like

You are loosing too much details in this ART version.

I disagree. I’ve just not introduced to many artifacts.

You may disagree, but look at the proposals here… The look of your image is flat, missing lot of details compared to the other proposals. And the artifacts (luma noise) is better to keep to keep some fine details and if you print you’ll see that the artifacts are not visible finally (this was talked recently, maybe in this thread, don’t remember).

1 Like

Others have added more noise on top of the noise for sharpening purposes. I’m testing denoise properties. Not artificial sharpening.

1 Like

0P9A0410_01.CR2.xmp (24.1 KB)

1 Like