Hi Mark, thanks for the tip. I missed this this option as it is only available for Deep Prime processing, not for HQ and Prime. Now I tried this same image again and the effekt is - alas - none
Here is the complete lineup from left to right @4x magnification: my traditional processing, HQ, Prime, Deep Prime with sharpening and without.
To my layman’s eye the sharpening artifacts are more or less the same in all 4 dxo variants. For all DXO-DNGs I used the very same postprocessing in Darktable (no extra sharpening there).
I even looked at the JPGs straight out of dxo: the same halo. So this is not from Darktable or my raw processing.
Again, this is pixel peeping … seen as the whole image, the overall processing effect is increased clarity and sharpness.
Hi Andreas. Could you upload the RAW file here (or to a RAW hosting service such as Dropbox) and post the link here so that we can better analyze what the problem may be?
Hi Mark,
this is a download link for the raw file: HiDrive
This focus on my wife’s glasses comes because I previously had mangled the colours before by unwise processing steps in Darktable (e.g. remove colour fringing will remove red from the edges of all red surfaces). As she features in many of my pictures it comes out as a bad idea to decolourize her glasses all the time
I fear, this might be my last opportunity to write in this forum. Last time the forum sw said “as a new user, you have 1 entry left” (or similar). I do not know the customs here.
I’m not seeing the artifacts visible in your screenshots. Here is a comparison between the original RW2 file(on the left) and a DNG(on the right) exported from PR2 as viewed in Faststone image viewer.
What image viewer are you using to make the comparison? Is it Darktable? I suspect that it is the image viewer that you are using to view the image that is causing the artifacts. Maybe you should try Faststone. It is free for non-commercial use.
Hi Mark, I use plain old Gimp or gwenview or kphotoalbum to view the files - nothing fancy. None of these does any processing (before I apply any in the case of Gimp). The JPEG from DxO is too dark, therefore I did some brightening (give the curve a dent to the left). This does only global changes and no local filters. To be sure, I started up my virtualbox again and did a fresh export as JPG.
In Gimp, I cropped and scaled by 400% and then added a screenshot and a brighter version. The middle is the brightness. The scaling is done without interpolation this time, so you can see the original pixels - just larger. I still see a 1px Halo inside and outside the spectacle frame and a 1px cajal line along the cheek. The slightly smaller insert is from my regular processing in DT (also scaled up by 400%). That has a few JPG artifacts but no halo and no cajal.
Best regards, Andreas
I’m not quite sure what to say. I can clearly see the aberrations that you mention in your screenshots. I am not seeing the same aberrations in my test of your file from PR2. BTW, my test had the Global sharpness set to full in PR2. So the aberrations are somehow occurring in your system but not in mine.
My only guess is that something in your system is applying additional sharpness or some other processing to your file either before or as you are viewing it. I realize that the viewers that you are using are perfectly adequate, but have you Tried Faststone? Just for fun, please give it a try.
Hi Mark, I see that it is the worst of bug reports if you cannot reproduce
I am quite sure, that the RW2 is right from the card and that I did nothing “wrong”.
However, you have a point:
I was running DxO inside a Win10 virtualbox on Linux machine. This VM has 16GB Ram and 4 CPUs and seems to be not particularly slow (~1 min per image with Deep Prime). As my PC just has the Intel integrated graphics, there is no GPU power to speak of. The native CPU is from the skylake family.
As AI algorithms are not really transparent, it might be that it performs differently (not only in speed, but in results) on different hardware. Just a guess.
For me it was an interesting experiment and I learned quite a bit on the side (looking somewhat deeper into my own raw-dev process in Darktable). So my pictures will come out a bit better in the future. Even without using DxO.
Thanks for the patience.
Cheers & Bye, Andreas
I wasn’t implying that you were doing something wrong, just that something is different between your system and my system.
I can clearly see the aberrations that you cite in your screenshots, so they are real and are caused by something. I just don’t know what is causing them. Maybe @Wolfgang has a point about camera resolution, but I’m not seeing it in my test. By the way, my test screenshot is also at 400% with the global lens sharpening in PR2 set to the ON position which is equivalent to 1.00 on the global sharpness in PL5. If it looks smaller to you, then click twice on my screenshot to see it at full resolution.
Anyway I’m glad that you are getting good results with Darktable, but please stick around, maybe someone else will be able to reproduce your results and shed more light on the problem. Don’t give up!
Q: Is this a smart move to modify one module if the request is fairly general?
A: Yes, if the module has an unreasonably high (plain text: wrong) sharpness correction.
A: No, because it does not allow to reduce sharpness in all other cases.
@Marie I have tried reworking @Andreas_S file in PureRAW and in Photolab 5. Here are the results: Leftmost is original RW2 file, center is PL5 DNG and rightmost is PR2 DNG, the latter two with Deep Prime NR applied and exported, as viewed in Faststone IV 7.5 at 700% magnification.
I can clearly see the halo Andreas describes in the PL5 version. For the PL5 version I reduced Global sharpness to -0.50 and the PR2 version is with GS set to on, which I believe is equivalent to GS at 1.00 in PL5.
The PL5 version is clearly sharper than the PR2 version even with GS set to -.50 in PL5 and on in PR2. Andreas’s work was in PR2 and the halo is clearly visible whereas my PR2 version shows no such halo. All corrections in my PL5 version(other than GS) were identical to the corrections made in PR2 and no other corrections were made. What is the problem here?
And what for users who don’t have PR but have PL and need to export to other application when needed to do things photolab can’t do ?
This halo in PL export rrblint shows is problematic.
In PhotoLab, we can tune sharpening with the respective sliders.
In DPR, it’s either on or off. Either/Or is always a bad choice, lots of grey between black and white!
Hi Mark, I would call this progress
As others have noted, my image in itself is not overly sharp. It was shot with a diminuitive GM5 and all details shown here are magnified a lot.
The halos are (for me at least) only a side aspect - albeit a significant one. But observe the colour change in the spectacle frame: the oversharpening (middle version) darkens the whole frame. The right hand version pleases me. At normal size it would certainly be sharp enough.
So, now the 100000$-question is, how to get this kind of result out of DxO PureRaw2 (on my setup).
I fear, that my current test version will run out first before there is a fix. But I’ll be happy to test again.
Andreas
… seen as the whole image, the overall processing effect is increased clarity and sharpness.
