[thix2112]

If anyone is interested here is what I started over at Adobe:

Re: Denoise AI and Pentax K3 mk iii DNG files - no... - Adobe Support Community - 13769456

[UncleVanya]

Pentax is far more aggressive, at iso 400 or has a 1 stop advantage over a full frame sensor. Without any trickery an apsc sensor should be one stop behind (ballpark). So canon might apply some raw denoise, but Pentax is at a physics-defying level. K-3iii exceeds the performance of an ideal FF sensor according to that metric.

Software like Toapaz would be less sensitive of course because it is designed to work on finished RGB images of any sort, it is in a certain sense more generic. And more likely to yield bad results on any other camera as well, it is very image dependent. On the contrary if anybody has used DxO Deep Prime or XD he knows that on most cameras it is absolutely rock solid. You can set it at 20% (to be safe) at any iso, always on, without ever looking at it, and the likelihood of getting best-in-class results is extremely high. Unfortunately XD is broken for the K-3iii. But it works acceptably (at least improves on DP, and the slider does something) on other Pentax cameras like the KP, so it is definitely possible to fix it to an extent.


https://photonstophotos.net/Charts/PDR.htm#Canon%20EOS%205D%20Mark%20IV,Cano...ax%20K-3%20III

That ignores that the k1ii also shows this effect.

https://www.photonstophotos.net/Charts/RN_e.htm#Canon%20EOS%20R3_14,Pentax%2...20K-3%20III_14

The gains with the k1ii are slightly greater than the R3 at high iso but not as much.

---------- Post added 05-04-23 at 08:22 AM ----------

The AI denoising algorithms need un-demosaiced data.

The ones developed so far? Or are you suggesting it’s a hard limitation that applies to future development?

[gatorguy]

Pentax is far more aggressive, at iso 400 or has a 1 stop advantage over a full frame sensor. Without any trickery an apsc sensor should be one stop behind (ballpark). So canon might apply some raw denoise, but Pentax is at a physics-defying level. K-3iii exceeds the performance of an ideal FF sensor according to that metric.

Yet with no apparent, aka identifiable, detrimental effect on the images? Pentax delivers better than expected performance IMHO.

What seems most confusing is proclaiming it a big positive for DxO DeepPrime XD, "The amount of detail that it is able to salvage is staggering" and "best in class performance, but you simply cannot accept the possibility that the Pentax K3III hardware might also deliver "best in class performance", and with no loss of detail? Camera hardware cannot do this? On the surface it doesn't seem like sound logic.

The same question applies to Canon's unavoidable baked-in raw noise reduction, employed more aggressively than the Pentax with it at all ISO levels, unlike the K3III which kicks in only above 200, or the K1II above 600 (?). All available evidence points to all three cameras delivering near best in class performance. I think long-time professionals shooting thousands of images a year might have noticed by now if these cameras were smudging detail like the Sony star-eaters ( A7II) from a few years ago.

But all that is beside the point made by the OP. If DxO can tweak the software to work better with K3III files it would certainly be appreciated by a number of our members. I've used their software for years and quite like some of the features it adds to my arsenal.

[simon_199]

.

What seems most confusing is proclaiming it a big positive for DxO DeepPrime XD, "The amount of detail that it is able to salvage is staggering" and "best in class performance, but you simply cannot accept the possibility that the Pentax K3III hardware might also deliver "best in class performance", and with no loss of detail? Camera hardware cannot do this? On the surface it doesn't seem like sound logic.

It could, in principle. In practice, it does not, because unfortunately software is not designed specifically for this Pentax camera that has a very peculiar raw file, and do not work well. At least DxO does not. So the overall final image quality is negatively impacted because I cannot take advantage of best-in-class noise reduction software. The jpeg output on the other hand is good, and probably better than others because the camera hardware does a fine job. That is the issue, the balance between the amount of noise reduction that gets baked into the raw and the amount that results in the in-camera jpeg. The sun of the two parts is fine and maybe even better than the competition. Unfortunately I am not interested in the jpeg output otherwise I would be extremely happy. It could have been very simple to just set a different balance, less aggressive in the raw, same total amount in the final jpeg. As canon (and others) are doing.

[gatorguy]

Unfortunately I am not interested in the jpeg output otherwise I would be extremely happy. It could have been very simple to just set a different balance, less aggressive in the raw, same total amount in the final jpeg. As canon (and others) are doing.

We actually don't know the details of what any of them are doing. They don't disclose it. All we have are results and almost universally users and spec-blogs seem to be pretty happy with them.

Regarding the K3III, I also only shoot raw as you do, only a handful of JPEGS in the years I've owned Pentax cameras with image accelerators, the K-70, KP and K3III. I've yet to identify any negative effect on photos so until I do it's not anything I have the least amount of concern over. My typical processing programs work fine with them, including denoising if needed. DxO I've been a bit less pleased with lately, but that's mostly due to support issues that I suspect will not be addressed. I'm slowly gravitating towards On1 instead of DxO which I used to heavily lean on.

How many years and millions of Fuji cameras did it take to finally convince DxO they should dedicate a little time to supporting them? I wouldn't hold my breath waiting for them to get around to this one. Best option is learning to get the most out of other very good programs that will process our K3III files across all ISO's more reliably. DxO DeepPrime in general stills works fine with my camera AFAICT, but I don't use XD, nor do I even use DeepPrime as often as I once did.

[simon_199]

I guess it boils down to personal preference, the level of pixel peeping that one engages in (I probably do way too much), for me the image quality I get at very high iso (6400+) out of dxo + the K3iii is not satisfactory on most images. Is the problem easy to ignore ? Actually yes, because I rarely push the uso so high. So we could agree that "in practice it is not relevant" ...but it depends on one's personal definition of "practice".

[gatorguy]

I guess it boils down to personal preference, the level of pixel peeping that one engages in (I probably do way too much), for me the image quality I get at very high iso (6400+) out of dxo + the K3iii is not satisfactory on most images. Is the problem easy to ignore ? Actually yes, because I rarely push the uso so high. So we could agree that "in practice it is not relevant" ...but it depends on one's personal definition of "practice".

Well stated.

[thix2112]

Adobe as opened an investigation on the ineffectiveness of its Denoise AI process on K3iii DNG files using a couple of my DNGs that I provided to them. Thanks Adobe.

[simon_199]

Adobe as opened an investigation on the ineffectiveness of its Denoise AI process on K3iii DNG files using a couple of my DNGs that I provided to them. Thanks Adobe.

Interesting. From samples I have seen the new Adobe AI denoise seems at least competitive with DxO. Even if the workflow seems cumbersome to me. As far as I understand is not fully integrated but requires to physically write an intermediate dng file or something like that. Correct me if I am wrong

[thix2112]

Interesting. From samples I have seen the new Adobe AI denoise seems at least competitive with DxO. Even if the workflow seems cumbersome to me. As far as I understand is not fully integrated but requires to physically write an intermediate dng file or something like that. Correct me if I am wrong

I have DXO Pure Raw 3 (which has a bit more adjustment options than v2). I think DXO works better on K3iii DNGs than Adobe AI, particularly when you set DXO Pure Raw to the DeepPrime setting (not the DeepPrime XD setting as mentioned earlier).

DXO will work directly within Lightroom as an external process and you can set up how/where the DXO raw file places its final output file inside of Lightroom.

Tom

[SrMi]

<snip>


The ones developed so far? Or are you suggesting it’s a hard limitation that applies to future development?

The ones developed so far require RGB data. Adobe could adapt its algorithms, but the target audience may be too small. Also, the noise in monochrome is much less of an issue than in Bayer/X-trans output. I do not think DxO will ever add it (based on communication on DxO forums).

---------- Post added 05-04-23 at 10:00 AM ----------

I have DXO Pure Raw 3 (which has a bit more adjustment options than v2). I think DXO works better on K3iii DNGs than Adobe AI, particularly when you set DXO Pure Raw to the DeepPrime setting (not the DeepPrime XD setting as mentioned earlier).

DXO will work directly within Lightroom as an external process and you can set up how/where the DXO raw file places its final output file inside of Lightroom.

Tom

Adobe's AI Denoise has a bug that prevents it from working on Pentax K3 III and K-1 II raw files.

[thix2112]

The ones developed so far require RGB data. Adobe could adapt its algorithms, but the target audience may be too small. Also, the noise in monochrome is much less of an issue than in Bayer/X-trans output. I do not think DxO will ever add it (based on communication on DxO forums).

---------- Post added 05-04-23 at 10:00 AM ----------



Adobe's AI Denoise has a bug that prevents it from working on Pentax K3 III and K-1 II raw files.

Hopefully Adobe will take action to fix it. They did respond to me and using DNGs that I provided. We'll see.

tom

[Wingincamera]

Hopefully Adobe will take action to fix it. They did respond to me and using DNGs that I provided. We'll see.
tom

I found that Adobe's AI Denoise works well with my KP's after I altered my processing. If I do any processing first, including sharpening, it gives poor results. But if I do Adobe's AI Denoise first without any processing, including no sharpening, it works fine. I then process and sharpen the new DNG file that is created. I find that Topaz Denoise's results are about the same as Adobe's AI Denoise, but then you have a TIFF file. Both programs produce new files that are about twice the size or more of the original files.

[simon_199]

I have DXO Pure Raw 3 (which has a bit more adjustment options than v2). I think DXO works better on K3iii DNGs than Adobe AI, particularly when you set DXO Pure Raw to the DeepPrime setting (not the DeepPrime XD setting as mentioned earlier).

DXO will work directly within Lightroom as an external process and you can set up how/where the DXO raw file places its final output file inside of Lightroom.

Tom

I have tried with the "new" DPreview raw samples at lower ISO and the observations I made a few months ago are unchanged. Actually DPXD works perfectly up to iso800. Perfectly="as advertised", i.e. less or same amount of residual noise in the smoother or out of focus areas, and improved details. At iso 1600 things start to break down and XD is significantly worse than DP standard. Artifacts appear in the detailed areas and the noise in OOF areas actually increases, and there is no amount of fiddling around with settings that makes it better. Actually as the ISO get higher, the result seems to become increasingly unresponsive to any slider for DPXD. The transition is not abrupt, that is between the full stops 800 and 1600 it transitions gradually from "working as advertised" to "broken". At ISO12800 DPXD is unusable, results are garbage. DP standard is acceptable, but below expectations (i.e. below what I get out of similar apsc sensors, including the KP). I guess it could be fixed but I can't estimate the amount of man-hours needed....
DP standard is quite good up to iso 4000-5000.

To sum it up what I do with my K3iii files is:
- up to ISO800-1000: DPXD, great results
- 1000-5000: DP with "noise model" slider turned down a few notches, results slightly below expectations, but satisfactory
- 5000-25600: DP, quality of the results is highly variable, depends on the image and specific portions of the same image might turn out good or bad. XD is basically unusable in this range

All in all it's more than workable. If one does most of his shooting in the iso 5000 to 25600 range, maybe there's something wrong
Since XD seems basically bugged (high-iso not only results are poor , but the sliders don't do anything, it is clearly abnormal) I still believe that it is a type of software issue that could be fixed and unlock even more potential.

[thix2112]

To sum it up what I do with my K3 files is:
- up to ISO800-1000: DPXD, great results
- 1000-5000: DP with "noise model" slider turned down a few notches, results slightly below expectations, but satisfactory
- 5000-25600: DP, quality of the results is highly variable, depends on the image and specific portions of the same image might turn out good or bad. XD is basically unusable in this range

Good to know. Thanks
Tom