In What free software does denoising best I took up an offer to post a pair of low and high ISO shots and compare them across different denoise engines, so I’m comparing a test shot I took yesterday from the span of ISO 100-25,600 and ran them pretty much unedited through Darktable, Rawtherapee, DxO and Topaz AI.
As a disclaimer I want to say that I’m not trying to award any program as “the best” and I’m not trying to disparage any of these programs because I think they all do a fine job. Also I’m not trying to suggest that anyone’s personal decision on the degree of noise reduction is better or worse. I just want to share this to help people understand what each has to offer so they can choose which best meets their needs.
So here are the files I’m comparing. The base image is at ISO 100, and I didn’t see any need to denoise it at all:
I think the results are good across all programs, especially after further refinement, and the differences become less significant if you’re not cropping aggressively. I hope this helps.
Normally, I use Neat Image by removing color noise in the raw converter (since this works quiet good) and only pass luma noise to NI. For this test I passed everything to NI.
Aesthetically I prefer Rawtherapee’s look (noisy but monochromatic). I don’t like the overly smooth ones and DT has a bit too much color left (but still better than Topaz & DxO). It might depend what the image is for though.
You can not really compare the high iso denoise at default settings between darktable/rawtherapee vs. DxO/Topaz. The first two programs keep detail and leave a lot of noise while the latter apps try to get completely rid of the noise.
To compare the performance one would have to change the settings.
This pair of images was very informative to me as I too own a Canon R7. These images highlighted the fact that I have to treat my high ISO images so different to my low ISO images because some of the usual modules I apply by default such as shadow and highlights brings out the noise problem even more.
Here is my attempt at the 26000 ISO shot. I used raw denoise for this image, which I rarely use because it can soften the image detail too much. I also used three instances of of denoise (profiled) at default values. I applied some sharpening using the diffuse or sharpen module but limited its effect to the brighter sections of the image to avoid sharpening the shadow noise.
EDIT: I often shoot at 32000 ISO with my Canon R7. When I do I don’t expect to blow the pictures up poster size and it is more about the ambience of the scene that is important to me and the image. So I feel the look of this monkey at 26000 ISO is very acceptable for the type of image I would be capturing with those high ISO numbers. It was actually soul destroying for me to realise how much was being compromised at 26000 ISO compared to the 100ISO of the monkey shot here. But, a great photo is less about technical perfection and more about the subject, the ambience and general look.
Yes, you are, two instances at default.
I also used an alternative demosaic algo.
The settings for the parametric mask are to exclude the lightest tones from the treatment. I use this type of mask when i sharpen too.
100% crop, no downsizing. Two instances of denoise profiled, the first for chroma noise, the second for luma. Using LMMSE demosaicer also made a nice improvement on this image.
I’d go a step further and say that a fuller edit to adjust for color and tone is necessary for a complete comparison. For example, the Topaz image is sharper using the default settings, but you can easily match that with DxO with some modest post sharpening. Topaz is often over-sharpened even at minimum settings so I relinquish some control with that program, which is one reason I prefer DxO.
But adding edits makes comparisons more subjective, so I only stuck to default DT and RT settings. However I should have mentioned that I did adjust the denoise parameters for each program to bring out the best I could from each image.
PS - I like what you did with DT.
Actually, I was surprised that the image turned out so well at ISO 26,500 considering how little of the full sensor capability was utilized. I had expected the image to be a total wreck at that level, and other photos I took at those high ISOs came out fine as well. The R7 is a pretty good camera.
Yes, the composition and look is far more important. But I mostly shoot wildlife and frequently have to aggressively crop, and that’s where noise becomes more of a problem for me. It’s not just that grain becomes distracting, it’s the detail gets clouded over. That detail is important to me because part of my objective is to showcase the intricacy of a bird’s plumage, or the structure of an insect’s eye to get people to reconsider and appreciate the common animals that they take for granted.
I’m not hung up on backgrounds being creamy smooth, but I’d like people to see something about a plant or animal that they hadn’t considered before. They can’t get close enough to see that, so I do it for them. At least, that’s my approach.
And fortunately I haven’t seen any instances of these programs hallucinating or creating detail that didn’t exist… something that I’m very sensitive to. OTOH I’ve seen some truly bizarre things come out from generative fill, so I have little interest in using those applications.
