Skip to main content

On1 No-noise, first feeling (:thumbsup:) but sliders do not go far enough?

Comments

8 comments

  • Brian Lawson Community moderator

    NoNoise has to handle image file formats differently than the RAW files. It uses 2 different AI engines based on that. For a RAW file the noise is removed from the raw data before it is demoasiced to create the actual image. When working with an image file format it cannot do that, hence the two engines. It is also why it takes a bit longer to show the image once you've opened it.

    0
  • Ray Miles

    Jonas, What happened to all your images? I would like to see them, as your analysis is very interesting.

    0
  • Ray Miles

    Where NoNoise falls down is with very high ISO images (my maximum is usually 12800) and the colour noise resolves itself into green and magenta blotches.

    DeNoise AI has the same issue even in Severe mode. DxO Pure Raw copes best.

    However, NoNoise is in its first iteration and there will no doubt be improvements before it is incorporated in Photo Raw 2022.

    0
  • Brian Davies

    I am curious just how long it is taking for most users to open a raw file and an image file in NoNoise AI? Joris Mak mentioned "it takes a few seconds to open the file", but I am finding it takes a lot longer than a few seconds. I am finding it painfully slow to open files. Once they open, all adjustments are instantaneous and saving only takes a few seconds. The results, so far, have been fantastic and I am super happy with the final images.

    0
  • Brian Lawson Community moderator

    I’m able to open a 40MB file in less than 30 seconds.

    0
  • Bernd PALMER

    Once NN is launched in the background, I'm able to open a .DNG with a resolution of 6000x4000, 267MB in just under 30 secs.


    0
  • Joris Mak

    So, I had construction going on during the summer and was away for a while. Finally have my desk in some sort of working order again.

     

    This is a picture I mean where ON1 doesn't seem to remove enough from the edges of the image:

    It is a ISO 6400 image shot on a Sony A7m2, but in very low light and the white balance quite a way of (very blue / underwater style as you can see). So there are huge transforms done to the data to get the image where you see it now.

     

    But you see a lot of red color noise in the sides (pants and legs of the girl) while the black of the kid's shirt if fine.

     

    The default auto settings for DxO DeepPrime give this:

    You'll notice that it also has color noise there, but less. It might be that the vignetting-correction highlights noise that ON1 doesn't clean up, and DxO takes it into account. But the sliders in ON1 were maxed out so I couldn't use more noise reduction even if I wanted.

     

    Also, this is zoomed in on the kid's hat with ON1:

    And this is DxO:

     

    DxO seems to handle the transition from 'this I can handle to this is the edge of my limits' much more gracefully :).

     

    I want to make clear that I'm NOT WHINING on On1 in this case. The results are awesome. DxO DeepPrime is best in class and we're looking at limits here. So it's not at all meant as complaining. More to highlight areas that might need some improvement or ideas for further development.

     

    For instance, I like that DxO can do this in auto mode... which is really good. But getting a preview of what it is doing is pretty much impossible. What I'm trying to say is that I can _only_ use auto-mode most of the time, or you need to export an image and judge the result, make tweaks and export again, repeat...

    DeepPrime isn't so slow on my system (it makes more use of GPU power so even a non-so-fast GPU can do wonders now) but still that's for from ideal.

     

    On1 on the other hand sits there for a bit loading (doing the algorithm, determining what is noise and not I guess) but then when it's loaded, you'll get a zoomed in full preview with sliders that are very responsive. This is awesome for dialing it in!

     

    Specially images where the noise isn't so extreme as the images I posted here, fine tuning the noise-reduction and sharpening can yield better results than relying on DxO's auto-mode.

     

    As a nicer example, look at this image:

     

    On1:

     

    Dxo:

     

    Maybe DxO seems sharper here, but there is a good thing to remember: DxO's applied their own 'lens sharpening' (profiled sharpening), while on the ON1 version I tweaked it myself to a point where I liked it.. which is most often just a bit less crisp (or judged at 50% or 25% how the image would be viewed).

     

    This is a 16 megapixel old Micro-4/3s image, shot at ISO 6400. I'm stunned :). If you don't judge the sharpening, but you judge the amount of noise and the amount of detail left I think they're very close.

    0
  • Brian Lawson Community moderator

    Nice write-up Joris. I encourage you to send this info to tech support. You can send them the link to this forum if you want to avoid re-writing the whole thing.

    0

Please sign in to leave a comment.