When is there going to be smart noise reduction based on the amount of noise or how high the ISO per image?
1 Like
I donât understand your concern. PRIME, DeepPRIME, and variants (part of PureRAW and PhotoLab) are smart noise reduction tools. They apply noise reduction and detail enhancement as part of the demosaic process as needed, taking all things that affect noise into account.
1 Like
If you click on advanced it will show you that the noise reduction is default at 32, and detail 30 or something. It does not change the amount of noise reduction per image. It doesnât change from 32 strength for a certain ISO and then go to say 60 with a higher ISO it edits all the images at the same strength. Am I wrong? It doesnât change the strength of the noise reduction per image.
I think I understand. You want the possibility to have the default settings vary according to ISO. I think thatâs a good request! It can help avoid the problem of too little or too much smoothing.
2 Likes
Unfortunately, DxO PureRAW | Feature Requests still doesnât have any Vote button.
Your assumption is not correct. PL will apply NR according to actual need - it does not treat all images in the same way (even tho the ârelative strengthâ settings are the same).
Thatâs why itâs safe to apply NR to all images - PL will automatically determine which ones it deems to need more or less treatment than other images.
- The strength settings are provided so that if you donât like the result for any particular image then you can override PLâs algorithms to apply less (or more) NR.
4 Likes
No, John - youâre missing the point. Yes, itâs safe to apply NR to all images. But would you apply it at strength 40-60 to all images? I wouldnât. I use strength 15-30 for low ISO and more when there is more noise. I think DeepPRIME XD2s is superior in providing good results at default strength regardless of ISO, but not perfect - and not everyone can use XD2s.
1 Like
DP adapts to noise level, so donât worry about it too much.
Note that noise level may vary greatly across the picture due to shadow recovery and vignetting correction. Unlike ancient denoising technology, DeepPRIME will not âspoilâ image parts with low noise. You have to adapt to the subject rather than to ISO. Decide how much grain/texture/structure you want to leave. Typical things to take care of include faces, hair, skies (even at base ISO), fine geometric patterns. For example, for headshots I use typically Luminance=25-30, for most other things I use defaults (Lum=40,FD=0), but in very few cases I preferred 10 or 70.
Only extremely noisy pictures may require manual intervention. Strongly negative âForce detailsâ will make the picture smoother, but with âwaxyâ or âplasticâ look. Using strong positive values, you risk getting unwanted artifacts, especially in parts of picture with fine details, or large âtonally smoothâ areas where you might get varied green-magenta color casts. Sometimes you may have to fix non-smooth transitions from grainy to completely smooth parts.
It doesnât need to, since it uses different metric than classic denoising. Actually this was my concern when switching from LR5.7 to PL7.1, as I shoot a lot in very high and variable ISO, with up to 3000 photos per session. I have RTX4070 and i7 14th gen CPU, so I can afford using XD2s almost all the time. With old Lightroom I spend a lot of time tuning denoising per picture, with XD2s I spend far less time for that. It may take several months before you get a good âfeelâ of what settings will suit you in particular cases and which things to check. After a while, you will pixel-peep less and less.
2 Likes
I was shooting in a very dark church, the stage or what ever you call where the podium is was about ISO 3200 and the audience was much darker. I had to shoot the audience at ISO 20000.
The Noise reduction was set to 32%. It was not enough to rid the noise of the audience. I did a single noise reduction on the audience to see what I would need to set the strength to and I had to change it to around 60. So it does matter what you set the strength to. It still would be nice since there is all this AI stuff coming out that it determines the camera settings to apply different strengths of noise reduction when your running 400 plus images through.
This is certainly true in extreme cases.
But if XD2s at defaults produces ugly picture, you must be really in need to try to save it. As a rule of thumb, if default XD2s is ugly, I just delete the picture. I had only one such case when Luminance=100 had to be used and something like -50 Force Details, but the result was good only for private use. It depends on camera/lens combination and rescaling (which is perhaps a good topic to discuss). With Z8, which is a bit noisy (tradeoff for fast sensor readout), and leaves by default a lot of headroom in raw, normally I cap ISO at 12k (some cap it on 8000 or 6400 to be safe), and that with good contrasty lenses for which denoising seems to work better. With D780 I can risk shooting indoor sports at ISO 30k and still get usable results (if lucky), but that is also what I would call extreme and may require case by case setting. However, in extreme cases there is no clear correspondence between ISO and optimal denoising setting, as the latter depends more on actual exposure, shadow recovery, and vignetting correction, which together may raise âperceived ISOâ by many stops. The default XD2s setting tends to be a little bit too aggressive for headshots but it should not spoil too much. Just choose 20-30 for your portrait presets and 30-40 for anything else.
I think there are too many variables involved here to expect a universal setting now. We should wait few years until ML assisted by human judgement makes it better. On the other hand, we are near the limit of information we can get from the photon noise â light is noisy by nature. Just look at some high ISO photos of fabrics and try to correct the noise by hand â youâll quickly see how the denoising artifacts may arise 
1 Like
But wonât the amount of denoising based on iso vary from camera to camera? Iâm asuming that some cameras will handle iso noise differently to others.
AFAIK, DxO models each camera noise and DeepPRIME uses that knowledge. On the other hand, the photon noise is well understood and behaves like Poisson process, so that component is predictibly unpredictable.
DeepPRIME and DeepPRIME XD2S are AI based tools. They analyze the content of each image and provide appropriate denoising for those areas that require It. DxO doesnât just provide the same algorithm to each image depending on ISO or the camera body. It is much more involved than that. The adjustments sliders are primarily available for those who wish to fine tune the results to their personal taste.
This is why the DeepPRIME variants are the best noise reduction tools currently available. By the way, since they are an integral part of the raw demosaicing process that is also why they are only available for use with raw files.
Mark
2 Likes
Yes - I essentially agree, Greg;