I'm primarily a Linux user (at least on my personal laptops), trying to stick with Gimp, Darktable, and RawTherapee. They are obviously lacking in denoise capabilities compared to the commercial AI denoise solutions that Topaz and DxO provide. Any way I was looking at a Youtube video on denoising in RawTherapee and saw the following comment:
This seems counterintuitive to me. Has anyone tried this procedure for noise reduction and has it worked well?
You can try it yourself. Downloads here: upscayl.github.io/
On virustotal.com, one vendor flagged it as malware, but all other vendors (many!) said the DMG is clean. For Linux it is an AppImage, same virustotal results.
Just gave it a try. In the past, I have noticed that this process works in using Topaz's Gigapixel AI where I upsized it and then downsized back to the original size. Note: Upsizing first makes more sense than downsizing first in that when downsizing, detail is lost that you don't get back when it is upsized to the original size.
But the limitations of upscayl are the types of files it can work with. It did not allow me to give it a RAW image, and even didn't work with a TIF file. But it did work with a JPEG file. I really don't do anything with JPEG files except the very last step to make it available to share.
This is what I was interested in, not the specific software. Thank you.
I did a bit more playing around with upscayl and I find it does quite a good job of up sizing by 4x. You do need to use some other program to down size back again. And it does remove quite a bit of the noise, but not as much as I can remove with other software.
You are given some choices on the upscaling type: General Photo (Real-Esrgan), General Photo (Remacri), General Photo (Ultramix Balanced), General Photo (UltraSharp), and Digital Art.
Not knowing what Esrgan meant, I googled it. See: arxiv.org/abs/1809.00219
Also, not knowing what Remacri meant, Googled it too. Sounds if this method of resizing can produce some artifacts.
I have only tried Ultramix Balanced and Ultrasharp. And the results do look like a very good resize.