Any ideas on why DeNoise-AI seems to struggle with the noise in this image? Perhaps outside what it is currently trained for?
Note: the posted image is the Before image for reference purposes (ie not processed with AI-DN).
The lack of good low frequency color noise reduction was one of the first things I noticed. I hope that gets fixed. In the meantime I am doing a second step with the legacy DeNoise which has very good color noise reduction. DeNoise AI seems to be oriented only toward high frequency luma noise, which is only part of the overall noise issue with many photos.
For high noise pictures Denoise requires setting it to maximum. The picture still shows some noise (though not bad) even then. I ran it through my custom preset in Studio which uses AI Clear and its much better. One problem may be that the skin tone color is not good (red) so it may confuse the color noise with the actual skin tone. I tried some color correction before running the picture.
Custom Denoise preset using AI Clear:
I agree about color noise reduction. I like the AI idea but prefer the individual noise reduction channels in DeNoise 6. I did some comparisons of the 2 and AI seems to sharpen too much even with enhance sharpness and restore details at 0. It also darkened shadows.
I agree, sharpening is too much even at zero, and shadows lose not only noise but also detail.
The lower photo of the saloon guy in Artisan West’s post, to me, makes his skin look plastic and too smooth for what someone that guy’s age would realistically look like… Unless your image processing goal is to create an illustration vs a more realistic photograph.
The upper image of the hat-wearing dude looks more like a realistic photo & realistic skin for a ‘middle-aged’ man - but it is softened a bit. Do you have Topaz Studio to try using a little bit of Precision Detail and possibly a hint of Precision Contrast to crisp that upper image up a bit. That would be the best if you can eliminate the softness of the upper image…
Here’s what I get if I do what I’ve suggested to the image maker by using the upper image of the saloon dude as my starting (for purposes here, Original) image then customizing sliders in Precision Detail and Precision Contrast within Topaz Studio.
Upper Image of Saloon Dude - Post DeNoise AI (snip from Artisan West discussion thread):
Artisan West’s DeNoise AI image (immediately above) brought into Topaz Studio - with Precision Detail & Precision Contrast adjustments added:
I consider products like DeNoise AI & Sharpen AI more like ‘functional’ software tools to use before getting into the “creative” software tools Topaz (and others…) offer. Normally I don’t think of them as the end, but rather a means to an ultimate image end.
p.s. I agree with Artisan West that the guy’s skin is very red in the original shot. The Studio processing helps tone that down a bit. As you can see in my lower image here. But, if it were my photo, I’d drain even more red saturation out of the shot - because the background is running red also. Was it the lighting in the scene? Or, are the shooter’s camera specs set to shoot warmer (some cameras can do that - I turned it off on one of my cameras)?
Bottom Image in My Post - Plus Less Red Saturation Overall:
Whoa! Did it do that color shift? Plus, accentuate all the color noise? Yikes. Is the product supposed to work on Raw images? Is that in the current product specs? I hadn’t tried it if so.
wowie. as they say…
So I just tried a proprietary raw (Canon) last night and it worked pretty well. On par with other jpgs or tiffs. I was all mentally set for it not to work after your experience. And, figuring I’d have to do a dng conversion for DAI to work with the raw. So it now makes me wonder if the program still hasn’t “learned” or tuned itself to work with certain color ways or combos of shapes (or something of that ilk). It might be a good image for the developers to feed to the AI learning algoithm… Because, technically, it can work with proprietary raw images.
Don, Kathy, etc what is best way to get a ‘stump the lump’ image to the engineers?
AIDN seems to struggle with images that have high color noise. It’s not clear if certain Raw files are affected more than others; I haven’t tested that. Perhaps it’s not trained for that, yet.
I also have problems with bad color noise removal with DeNoise AI. My workaround is the use of Lightroom or Photoshop to remove only the color noise and use DeNoise AI only for the luminescence noise removal. This work ok, but In future the color noise removal should be part of DeNoise too.
In my opinion the color remove is too limited and worse than Lightrooms color remove.
Do you have example images you can post (orig & after DAI) for the engineers to use to ‘tune’ the algorithms?
Yes did upload some.
Like this one for show the better use of Lightroom plus DeNoise AI.
Only DeNoise AI
Lightroom only color noise removal rest DeNoise AI, normaly the setting could be lower much more to get better result.
At ISO 100K, DeNoise AI is without help from Lightroom useless.
DeNoise AI alone on my filtering
With Lightroom color removal, again the filtering could be lowered only this high for comparsion
Do you use DNG (compressed) or Canon MRAW files?
Maybe the 6d Mark II file format isn’t supported right now?
Just attempted to use Denoise AI on a very noisy image and it did little if anything for the color noise. Had to run it through Denoise 6 to use Clean Colors slider to fix.
Just way too much about Denoise AI that still should be in beta, including a redesign of the UI to match Studio and other plugins.
If you have it, use AI clear in Studio. It does well on luminance and color noise.