DXO PL 5.4 / ON1 22.5 trial - noise reduction test

I made a test with PL and ON1 noise reduction.

Image is from: Nikon D850 Sample Images | Photography Blog

Loaded in ON1 22.5 and PL 5.4

Processing:
quick and dirty grading to get a slightly flat image and to let the image look similar in PL and ON1 (I was suprised how different the processing results of PL and ON1 are at default settings).

Screenshots taken from XnView image viewer set to sRGB and 100% (I am watching on an sRGB calibrated monitor).

Export to tiff with NR switched off in develop settings:



Interesting: The ON1 image looks like there is croma NR going on even though NR is switched off.
Cross check: Classic HQ NR in dxo activated, croma set to 80, all other options to 0.


Now dxo and ON1 images look similar again.

dxo DeepPrime and ON1 NoNoise at default settings:





For me at default settings this is a win for ON1. DXO has less noise but lost way to much detail. On the other hand, for me ON1 is overdoing the sharpening / detail preservation.

Edit: First conclusion: For me DeepPrime and ON1 at default settings are both not suitable for this image.

DeepPrime (+ some quick n’ dirty unsharp mask) and NoNoise at custom settings:
Edit: (check also the third batch: there I use Lens Sharpness not Unsharp Mask which I think gives better results)





18_nr-custom-ON1

My conclusion:
Both are great. Is there a “winner” for me? Not really.

One note though about editing.
The result of the unsharp mask depends heavily on the result of DeepPrime (if there is noise, the unsharp might increase it). So one need to judge the result of the unsharp mask in the tiny DeepPrime preview. For me this was a challange. In the end I had to do exports to get an impression of the result. I would prefer to have a feature to at least make the preview bigger (even at the cost of longer preview processing time). ON1 is better in this regard because it shows the NoNoise in the normal edit window.
On the other hand for me dxo runs way smoother and quicker than ON1 (tested on notebook, Core i7 10th gen, 16GB, RTX 2070-maxQ with 8GB vram).

4 Likes

Your tests are flawed. NoNoise contains fairly heavy sharpening in it’s default setting, Deep Prime does not include sharpening. Try turning on Lens Sharpness in PL5 at default setting and turn off the Unsharp Mask then retest.

3 Likes

I agree and wrote about the heavy sharpening of NoNoise. That’s the reason the second patch shows NoNoise with sharpening and detail at zero not default:

Here comes the third batch with dxo Lens Sharpening instead of Unsharp mask and NR at 16 instead of 13.

Please watch in 100% zoom (click once on images).

25_nr-custom_B






Thanks for the advice to use Lens Sharpness. It works better than Unsharp Mask.

I also tested Topaz and it did a great job, but dxo and ON1 are so close that I don’t see a benefit in adding Topaz as an additional workflow tool.

Okay, we now know that, applied to this very image, PhotoLab and ON1 can reduce noise while leaving details intact to about the same degree.

Both apps seem to do what is technically possible now. :ok_hand:

1 Like

@cgidesign

‘comparing’ your last screenshots, to me it looks as if the (in focus) DxO examples show a just slightly better plasticity (better 3D) – watching in AdobeRGB as well sRGB

1 Like

Maybe.
I am starting to get an understanding of the DP-NR and the more I test it, the more I am impressed.

Here is another side by side.

This time DP-NR set to 16 instead of 13. Maybe that’s already a bit to much as the skin looks quite smooth. But anyway, Lens Sharpnes increased from Global 0 to 1.3 etc. and I added a bit Fine Contrast.


It’s subtle but the image got more crisp even though it has stronger NR. Maybe that’s what you mean with “it looks more 3D”?
Really impressive what’s possible.

1 Like

And a last test set. I am still impressed. Comparing the ISO 100 with the NR treated ISO 25600 leaves me speachless.

The image (from Nkon D850, Nikkor 50mm f/1,4G)

First: ISO 25600 without NR vs. ISO 100:
(no lens sharpness)






Second: ISO 25600 at NR-DP default vs. ISO 100:
(no lens sharpness)






In some areas the loss of detail is obvious in some not. But in general, getting this result out of an 25600 ISO image is absolutely stunning for me. I dare to say, an average spectator might not even notice that the image was made from a high iso source.

No comparison with ON1 this time because ON1 has an issue with rendering this image. For some reason it decides to put a super contrasty effect on the alphacool logo even if every adjustment is off. I sent it to ON1 support and got the answer this might be solved in the next version.
Regarding this image DXO PL is the clear winner for me :slight_smile:

1 Like

For a real test at how good the NR methods do the tests with ISO 3200 or 64000. This is usually the point where NR shows its true colors. Especially with furry animals and birds and landscapes with lots of fine detail.

For my Nikon D7500, which I believe actually has the same sensor as the D850, this is the ISO point, at least in the past, I found Deep Prime to be a clear winner. I have not tried the ON1 stuff yet this year.

I general found once “tuned in” with lens sharpness and noise reduction setting (only did this once and save it as part of various presets), the Deep Prime results at ISO 3200 were equal to or better than other RAW processors at ISO 400 with a very good balance of detail vs. NR.

Keep in mind that DeepPRIME and PureRaw are intended for use with all raw images regardless of the ISO. It is rare to find an image at any ISO that DeepPRIME does not improve.

Mark

2 Likes

I agree with both of you. The test was about the worst case so to say.