Images seem very noisy while edited, but export to jpeg seems much better?

DPL 6.9, macOS Ventura 13.5.1, Fuji X-T5 RAW

I am editing a set of RAW images shot on a Fuji X-T5 at ISO 6400, 1/500th, and f/5.6 and I have noticed that when I am editing images, they are much noisier on screen then after I export to JPG. It is making it kind of hard to learn how to use the DxO Denoising options since the output isn’t matching what is on screen (to be clear, the JPG looks better).

Is this normal or is there something I am doing wrong?

Use the small loupe in the adjustment palettes to preview noise reduction. You will only see the effect of HQ noise reduction (not PRIME) in the full image preview. This is documented in the user guide.

This is normal and disappointing.

Only HQ denoising is visible in main window.
Other denoising modes are only visible when saving images or in the very very very small window in the denoising palette (which is so near to unusable).
So you have to save images every time you want to check your settings … Then choose, then delete images test.
Not really the best workflow …

This is not a only mac issue.

I think you might be more disappointed if DeepPRIME/XD were rendered into the main viewer, as every time you moved any slider you’d be waiting many seconds for the view to update.

There are two choices:

  1. Just use the magnifier provided in the 'DxO Denoising Technologies" panel and use the Magnifier tool to inspect different areas of the image as required.
    CleanShot 2023-09-01 at 15.27.35@2x

  2. Make your noise reduction adjustment first (probably using the magnifier as well), then export as “DNG (Denoise & Optical Corrections Only”, then you can do the rest of your adjustments on the resulting file.
    CleanShot 2023-09-01 at 15.29.03@2x

I’d sure love if DeepPRIME could be rendered in less than a second, but it can’t and having the magnifier is the next best thing. I’ve processed many thousands of photos this way without really worrying about it.

I would be less disappointed if I had CHOICE : USER CHOICE.
For example, preview window (in denoising palette) should be resizable and undockable.
An other one would be to select a custom resizable area in the main window where full processing would be applied.
There are many workflows and many hardware powers.

I have softwares which use extensively graphic cards. They allow to use graphic card powers of every wokstations on the same local network (or selected ones) when working on one workstation (I’m not talking about batch processing, but really when currently working on one item - they allow this in batch processing too, of course).

There are many solutions DxO didn’t explore.

1 Like

This does not solve the problem of tweaking denoising and sharpening tools.
For now, if you want to tweak it you need to save full images test, choose, then delete images test. Not a really an efficient workflow isn’t it ?
I suspect most people let everything at default values because too heavy to tweak.

Preview denoising window is absolutly too small to tweak efficiently those tools (denoising and sharpening). I would say it is near from not usable. It is only something like 1/1200 the surface of my images ! (I caculated this, but this number is from memory so maybe not precise - it’s between 1/1000 and 1/1500) . How to judge with 1/1200 the surface of my image ?
And worst, it needs to recompute any time I change and restaure an already tested setting (hard to compare).

I would need to be able to select important zone of my image, which can vary in size and proportion from one image to an other.
I would need the result of processing this zone to be cached automatically in memory (RAM, not drive) for several tests (any number I want) and be able to switch instantly those caches to compare. And have a free memory button to delete those caches when done.
And switching caches should recall settings used for displayed cache in the palette concerned.

They, maybe, I would save a DNG to process further adjustments.

All good suggestions, but not what I was responding to, which was your desire to have the main preview show noise reduction. At this time, every slider you move gives a near-realtime update to the main view. Yes something could be done to improve on what we have, but I don’t think even the most powerful graphics system is going to make, for example, DeepPRIME XD ‘near-realtime’.

No, which is why I just use the magnifier. As I said, I have processed thousands of images. Very many of those have some level of noise I want to remove. I just move the magnifier around to check key areas — details I want to ensure, flat areas I want no artefacts in — and almost always I do not need to reprocess after my first export.

If you can stick to DeepPRIME, as I mostly do, it’s a quite simple task. DeepPRIME XD is much more susceptible to creating artefacts, but still I manage with the magnifier and very occasionally I miss one area of these and have to revisit once.

I strongly disagree. Granted, I am basing my opinion on my own experience, but while it’s not ideal, it is far from being unusable.

When tweaking denoising settings, I tweak sharpening too. So for me both are linked in the same operation, so what I say about this so small window preview is for this “double” operation.

I do the same thing as you do, because I have no choice, and so yes, in some way it works.
Near from usable is maybe a little too strong, since I use it because no real other choice.
But not efficient at all and big time waisting isn’t too strong when tweaking those 2 process (denoising + sharpening) :
Doing screen grab, save them and compare them to tweak settings isn’t.
Having to do it several time to check several parts of image isn’t.

As I said, I suspect most people let every settings at default values because too heavy to tweak.

It is a bit disappointing. But not unlike Adobe’s implementation. Those two developers have chosen a very powerful algorithm that exceeds the real-time capability of today’s hardware. One day, when the hardware catches up, things will work more smoothly. In the meanwhile, we’re left to make do.

Photo RAW’s NoNoise AI is not quite as good as DxO and Adobe’s technology, but can be rendered in real time, at least on my hardware. It’s more convenient to use and seems to be “good enough” for the higher ISO (up to 12,800 from an OM Systems OM-1) files I’ve tried with it.

1 Like

Maybe. Or algorythms will evolve and be more and more powerfull and need more computation power as hardware evolve and things will stay at about the same level.
As we have seen since last 20 years in other domains that require intense calculations too.

Hopefully this time is coming soon, at least on MacOS since their Silicon Chips are getting more and more powerful.
In the meantime, even if it is not realtime, DxO could give us an option to generate a temporary “dng” file without noise while editing and delete it automatically. So we do not need to do extra steps to have a near realtime use… :thinking:

That’s what I do, now, but manually as my first step for a noisy image. But I don’t, and wouldn’t want, the .dng file to be deleted. When I’m finished editing, it is the .dng file that holds all my edits.

That’s why the process should be well thought and the edits saved in the “orginal PhotoLab” format like if we were editing a raw file normally.

I don’t use the defaults, but I do have defaults baked into the few presets I use for different types of images. I usually just stick with those for sharpness and tweak noise if necessary, based on the magnifier. Very occasionally, I will export a batch and when checking the final output I will find a one or two with “too crunchy” sharpening, such as with fine grasses, which need the sharpness to be moved back some. I am getting better at predicting when this is needed.

Which leads me to another thing. The sharpness of the final output matters not only on the sharpness of the ‘original size’ but also the output size. I do not export my images at full resolution but always export 6 megapixel. If the original is close to my full 24 megapixels, that can mean something very different than a heavily cropped one that’s maybe 10 megapixels before export.

As good as they are, they’re nowhere near crunching DeepPRIME on a modern RAW file in anything like real time, let alone DeepPRIME XD.

Sure, experience is the best value.
But being able to immediately, easily and very precisely (for pixel peepers) compare several tests should be consider as a must have when creating this kind of software.
Not only for denoise/sharpness but for final result too.

Absolutly, and what the image is intended for (what support will display it) matters too.