[Madaboutpix]

It is pretty clever of them to provide you with a fully-functional trial version. I've (re)processed too many DNGs to be able to ignore that the newer versions do look discernibly better on SmugMug and on my editing monitor. So, the point of no return is already behind me, and I know I will get the upgrade code, alright.

Also, with my time-honoured K-3 receiving these regular IQ enhancements via the PL upgrades, the waiting time until I will be able to afford a K-3 III gets a little easier to bear.

[rpjallan]

I'm not considering upgrading at the moment. I rarely shoot at high ISO & I can't really see anything else that would make me spend the money. Will keep an eye out for any upgrade discounts though.

[simon_199]

I'm experimenting with the new PL6 and the K-3 mark 3. Since I got the camera I was under the impression that the DeepPrime NR did not do as good a job as on my older cameras.
With the new XD variant I think that there is some evidence that support (that is, I guess, fine-tuning of the model for this camera model) could better.
I will post some examples from an ISO12800 test image. It was taken in "bad" light, but it is representative of the general impression I have in the range of about ISO3200+ to ISO12800+.
Up to ISO1600, perhaps 3200 both DeepPrime and XD seem to do a good job and behave "as intended" (i.e. similarly to other cameras).


This is shot with a Fuji XF10, 24MP APS-C sensor (non X-trans). Images are named with a suffix contaning the settings, the first number is the master (luminance) slider, if there is a second number I refer to the "noise model" slider that is new in DPL6.
As you can see at the default luminance of 40, DeepPrimeXD produces equal or slightly less grain in the blurred areas with no artifacts, and preserves even more detail.
And what's more, there is a strong dependence of the residual noise on the luminance value, from 0 to 100 (last two pictures) I am clearly exploring the trade-off between detail preservation and removal of grain.



On the contrary with the K-3iii at high iso at L40 XD produces way more grain in the out-of-focus areas, and there is almost no difference in going from 0 to 100 with the luminance slider.
This is a general behavior that I have observed at high-iso and is not strictly picture-dependent (this test shot is rather crappy).
Using the new "noise model" slider with negative values I can obtain results similar to DeepPrime (last picture) but more often than not this introduces an unnatural look in the out-of-focus areas and does not significantly improve detail retention with respect to DeepPrime.
Even keeping the "noise model" slider at negative values, there is very little sensitivity to the "luminance" slider for DeepPrimeXD. This is unlike the other cameras that I have tried.
On the positive side, at lower ISO this new parameter seems to work well and often allows to find a better tradeoff with respect to DP. Furthermore at lower (say <= 1600) ISO, also the sort of inability to remove grain in the blurred areas is not there, it seems to work as intended and really improve (albeit slightly) on DeepPrime.



Any thoughts ? Has anybody found the same issue ?

[Madaboutpix]

Any thoughts ? Has anybody found the same issue ?

Don't own a K-3 III, and therefore have no high-ISO DNGs to test on. If I experienced odd behaviours like you did, my first tendency would be to check on the camera if I had disabled the in-camera High-ISO Noise Reduction. For DxO's denoising technologies to be most effective, they likely should be left undisturbed by extra noise reduction applied by the user in-camera (I'm well aware of the Accelerator Chip and what it does, but I understand it would be counterproductive to bake any more NR on top of that into the file prior to processing in PL). If I found that the odd behaviours persisted nonetheless, and other users reported similar experiences, I would raise the issue on the DxO Forum.

[simon_199]

Don't own a K-3 III, and therefore have no high-ISO DNGs to test on. If I experienced odd behaviours like you did, my first tendency would be to check on the camera if I had disabled the in-camera High-ISO Noise Reduction. For DxO's denoising technologies to be most effective, they likely should be left undisturbed by extra noise reduction applied by the user in-camera (I'm well aware of the Accelerator Chip and what it does, but I understand it would be counterproductive to bake any more NR on top of that into the file prior to processing in PL). If I found that the odd behaviours persisted nonetheless, and other users reported similar experiences, I would raise the issue on the DxO Forum.

The high-iso noise reduction settings don't affect raw they are only for jpeg. I had it on Auto, but raw files should be 100% unaffected. I believe that the accelerator chip could (indirectly) be the source of the issue as it makes Pentax files a sort of an "outlier" with respect to the "average" of raw files on which the baseline deep learning model is probably trained on. There is most likely some fine-tuning for the specific camera that might or might not include training a pre-trained model on a camera-specific dataset. If that's the case maybe there is room for improvement.

Interesting: if there is a DxO forum to report possible issues I'll give it a look.

Furthermore, if anyone not having a K-3iii wants to try I can share some raw files.

[c.a.m]

Any thoughts ? Has anybody found the same issue ?

@simon_199, thank you for your notes and sample test images. Very useful.

I have also been trying out PhotoLab 6. In fact, I have bought the package as an upgrade to PL5 ('birthday present' ).

I have been playing with RAW files from my own K-3 Mark III, as well as older files from my K-3 II. I've been using mainly the ISO 12,800 PEF files from the Imaging Resource's review of the camera (sample 'studio test' files are available here: Pentax K-3 III Review - Samples), because they include a variety of textures and details, and their methodology and shots are well controlled.

I have found exactly the same results as you, so my 'experiments' corroborate your findings. I agree with your results concerning the relative insensitivity of the Luminance slider. I haven't completed writing my notes on my investigation, but my general subjective conclusion at this point is DPXD can produce slightly better results than DP for RAW files from the K-3 Mark III. How much better? Maybe 10-20% in certain images.

In areas of images that lack details, for example blue sky or smooth walls, DPXD can introduce splotchy low-frequency artifacts or 'texture', unless care is taken to find the best NR settings. Like other noise-reduction schemes, it's necessary to balance the DPXD settings between smoothness and detail extraction. Other users have noticed similar issues, for example as discussed here: DxO PL5 vs DxO PL6 NR algorithm - #20 by zkarj - DxO PhotoLab Mac - DxO Forums, and a possible issue noted with K-3 Mark III images: https://feedback.dxo.com/t/deepprime-xd-doing-nothing/28759

In low-ISO images, I believe that it is possible to extract just a bit more detail compared to DP alone, probably because of the beneficial demosaicing step under the denoising scheme. So, every bit helps.

I am quite experienced with PL5 DP, and have developed a good understanding of how to set the Lum value under various ISO levels. However, I think that I'll need to experiment more with PL6 to understand the interplay between the settings for Luminance, Noise Model, and Lens Sharpness. Unfortunately, the PhotoLab User Guide provides scant information on this aspect of the package, and users are left on their own to discover the best settings for various image conditions.

I read your posts at the K-3 Mark III sample image thread. Also very interesting and insightful.

Like you, I suspect that the built-in 'accelerator' denoising scheme in the K-3 III may be the reason for the unique behaviour of DPXD for this camera. I also looked at a RAW file from the K-1 Mark II at ISO 12,800, and observed a similar behaviour of the Luminance values. Even at Lum=100, smooth areas in an image still show low-frequency splotches (for example, the plastic Proportional Scale in the Imaging Resource studio scene), albeit at less-apparent levels than seen in the K-3 III images. In images from other cameras, a setting of DPXD Lum=100 appears to smooth out these areas completely -- they are void of detail.

- Craig

[ehrwien]

I am quite experienced with PL5 DP, and have developed a good understanding of how to set the Lum value under various ISO levels

And here I am, having never even touched the Luminance slider for the whole existence of PL5. I just applied DeepPrime to all images I developed at the standard setting. How much potential for cleaner files did I give away by that?

[c.a.m]

In PL 6, Deep Prime XD (DPXD) takes longer than DP to export image files. For example, to export an ISO 12,800 TIFF image on my desktop PC, DP took 23 s while DPXD took 36 s, or approx. 50% longer. (Dell XPS 8930, i7-8700 3.2 GHz, GPU NVIDIA GeForce GTX 1060 6 GB)

I have updated the driver for my GTX 1060 to the NVIDIA GeForce Studio Driver 522.30, dated 18 October 2022. The release notes indicate "This NVIDIA Studio Driver provides optimal support for the latest new creative applications including Topaz Sharpen AI and DXO Photo."

The new driver certainly improved processing speed, reducing a typical DPXD TIFF export from 36 s to 22 s. The time taken to render a full-resolution chip in the PL6 magnifier window when viewing denoising action is also significantly faster.

- Craig

[c.a.m]

And here I am, having never even touched the Luminance slider for the whole existence of PL5. I just applied DeepPrime to all images I developed at the standard setting. How much potential for cleaner files did I give away by that?

Oh no, now you will need to re-process all of your images, experimenting with the slider... (just kidding).

Actually, for my K-3 II and K-3 III, the Deep Prime default of Lum=40 seems to do a reasonable job for most images -- it gives a good general balance between smoothness and noise. In images at higher ISO levels, one might be able to squeeze out a bit more detail by lowering the Lum value, but I don't think there will be a noticeable difference except when viewing images at full-scale (100% zoom).


- Craig

[ehrwien]

I have updated the driver for my GTX 1060 to the NVIDIA GeForce Studio Driver 522.30, dated 18 October 2022. The release notes indicate "This NVIDIA Studio Driver provides optimal support for the latest new creative applications including Topaz Sharpen AI and DXO Photo."

The new driver certainly improved processing speed, reducing a typical DPXD TIFF export from 36 s to 22 s. The time taken to render a full-resolution chip in the PL6 magnifier window when viewing denoising action is also significantly faster.

Wow, do you know how stupid I feel now? I have the very same GPU, and I've not updated the driver in ages because I thought "It's an old GPU" and "Never touch a running system"... a K-1 file that took me 40 seconds to export with DeepPrime now takes less time (~3/4ths?) with DeepPrimeXD. Why didn't I even remotely consider that the GPU driver could be optimized for the software I use?

Thank you!

[Madaboutpix]

Furthermore, if anyone not having a K-3iii wants to try I can share some raw files.

Yep, that must be me. Would be really cool to have some DNGs to study. After all, the K-3 III will likely be my next camera, and I would want to be confident that its DNGs develop well in my raw converter of choice.

Interesting: if there is a DxO forum to report possible issues I'll give it a look.

By all means do. And do report or confirm the problem. DxO may take a while to react and come up with a solution, but eventually many issues get fixed in updates or new versions.

[UncleVanya]

Check KP files also - and K-70 - if those are similar that would point to the accelerator.

[Larrymc]

For me the PL6 Jury is still out. I'm not seeing much improvement in the file output from DeepPrime vs DeepPrimeXD for my Mk III images. I've got 22 more days on the trial and I just updated PL6 to version 6.0.1 Build 33, so I'll decide by then.

[simon_199]

Thanks for the replies and for confirming my findings. In the topic on the DxO forum referenced here it seems that somebody has already reported the issue but I'll maybe open a dedicated thread and share the same images I posted here just to increase the visibility. I'll also check files (from the usual studio samples) from other Pentax cameras with Accelerator Chip. I expect something similar but not necessarily as bad. I have to say that I am under the impression that also the regular deep prime is not as effective on such cameras, by comparison with other cameras with similar sensors, but that is much more subtle and difficult to prove. I will also post here a couple of raw files to try.

Furthermore thanks to another user here on the forum I have found out that pixelshift images (at least from the K33) are not properly supported, even in PL5, and for that I have opened a support ticket since I own the software. We'll see how it goes.

[simon_199]

I have opened a topic on DxO forums, still pending, will let you know. Meanwhile I have tried a KP file and to my surprise it works surprisingly well:



Regarding K-3iii raw samples there are a few galleries available on the web for you to try. Playing around with those pictures I'm a bit puzzled. The DpReview do not seem to exhibit this strange behavior whereas the imaging resources pictures do...
Perhaps it depends on the lighting, or the lens, or some obscure camera setting (it's not the RAW high-iso noise reduction because I have checked with DCU and it was on "auto" for the DPR set).

Pentax K-3 Mark III sample gallery: Digital Photography Review
Pentax K-3 III Review - Gallery


This one (Digital Cameras, Pentax K-3 Mark III Digital Camera Test Image) for example, works terrible with DeepPrime XD even if it is just iso1600.