Maybe smooth ?
Soft - maybe. DxO’s XD2, when offered in PR4, was criticized in the forum as too harsh in the normal setting, and most folks preferred the “soft” setting. Also, most folks on the forum seemed to expect the XD2 algorithm in PL8. It’s obviously different.
Has anyone compared the PR4 implementation of XD2 vs PL8’s XD2s? I’m travelling, so hope to get a personal comparison in later, but would appreciate any thoughts.
Thanks!
I think I read they have updated pr4 with XD2s. Has to be confirmed.
Distortion correction moves pixels around and introduces some rounding errors. Sometimes, the pixel taken from position A will be placed in position B or C. Do that for a bunch of pixels near A and you’ll get bunches near B and C…and the result will look less sharp.
1 Like
Some reviewers have - for example:
There might also be quick comparisons in this forum, but you’ll have to search just as with the Internet as a whole. You can also do your own comparisons if you haven’t run out the 30-day trial periods for PL8 and PR4.
If PR4 now has XD2s, it isn’t mentioned in the release notes, which is weird and makes me think PR4 hasn’t received such an update yet. Perhaps it will, though.
Yes, since there is a deformation, there obviously are interpolations to find in between pixel values which can only lead to very very small loss of quality (generally unnoticeable).
This is why I enable it only when needed.
Put black stripes on the car’s body, paint the brakes in red … and you know it’s faster. 
@JoPoV – perhaps you should add “moire” to the title.
Take a look at dpreview https://www.dpreview.com/reviews/image-comparison for more raw examples with moire. Look at the lithography in the middle of the left half of the image, where moire (and maze) is most visible, but it can be present in other areas too. Choose from 20++mpx cameras with weak AA-filter. It seems to me that for higher ISO images DP deals better with moire than for low ISO, which one could suspect. The scene is artificial, so maybe someone in the forum can share real life raws, e.g. birds, architecture, weddings, or other events, with various types of moire?
Being DxO somewhat scientific-related, I would expect them to work on this. But maybe moire is truly deadly for everyone?
@JoPoV : If you have D850, I’m not sure you will find Z8 worth buying, even if Plena cannot be used on your camera. Perhaps wait for next generation of EXPEED and AF algorithms. Personally, I consider Plena to be a perfect universal, rather than a portrait lens. Depends on your taste and “shooting mode”. Now, not to derail the interesting topic, I’ll try to send you a more detailed PM within a week or so.
???
It’s the first word of the title.
XD did the best job on it (at least for colors, not pattern, as I describe above). Xd2s does not.
Oops, sorry. It was too pronounced
So far, I agree. But I would like to see more examples from the forum users, to try to find some “rules of thumb” for different types of moire.
Have samples with pattern, but I would need to search more for those with pronounced colors since pattern seems about equivalent with deepprime, XD and XD2s and only color differs (a lot : magnify image at 100% (forum displays a reduced image) difference is hudge).
I will look deeper in not keepers, but lot have been deleted.
But I had a bug with vc - maybe .dop corruption, and since I rely on .dop mostly (I like to organize my files with explorer, so database is probably not clean), I’ve slowed down the use of v8 while waiting for support feedback. I sent them a reproducible case.
No response since wednesday. Only acknowledgement of the ticket.
Here’s a test case that I’ve been using for about seven years. First, PL7 with DPXD, Luminance 30:
Next, PL8 with DPXD2s, Luminance 30:
While moire patterns persist with some false color, there are far fewer artifacts in the PL8 image. If you zoom in a lot, you can see that DPXD’s artifacts are crazy. Regular DeepPRIME (not shown) is actually worse in this case. DPXD2s gives a smoother result with no horizontal banding and less microcontrast in the shadows. Too smooth? I don’t think it’s actually losing more fine detail than DPXD at Luminance=30 (I usually prefer a lower setting). However, while crosshatching is greatly reduced there is stronger false color and contrast in the moire patterns with DPXD2s.
I set moire correction to 100% in both PL7 and PL8.
These images are a bit small, but the point seems to be clear. Do you have a raw to share?
Your example may lead to a pre-preliminary (stupid ?) “rule of thumb” – for moire, use DP for birds, and use XD2s for architecture. Ok, that sounds too simplistic…
PS. I just tried to get the moire with moire-prone Z8 and didn’t succeed. Any hints how to get one easily?
Sorry, I don’t want to share the RAW file. But here’s a crop from the out-of-camera JPEG for comparison:
I know these crops are small - just download them and zoom in more. I think they demonstrate the flaws in DxO’s demosaicing rather well - and for that reason I’ve shared the RAW with DxO numerous times. DxO far surpasses the camera in other ways, of course - and I’m glad to see this image being rendered better with each new iteration of DeepPRIME. But significant problems remain.
Your image is very similar to two images used as standard references in scientific papaers about demosaicking/moire, so surely DxO keeps an eye on this problem. But moire problems seem to be very hard to resolve by software…
Anyway, your and @JoPoV examples show that there’s no simple answer to the question which of DP variants is better.
Maybe I’m just getting too much worried with the Nikon D4->Z8 switch and possible moire problems?
One would expect so - but my interactions with DxO over the years leave room for doubt. If they do such scientific tests, the results will vary among cameras and will not necessarily be considered important to real-world use of the software. So our feedback through support.dxo.com and beta testing is important IMO.
Another consideration is that the DxOMark review of the Z8 makes no mention whatsoever of moire: Nikon Z8 Sensor test - DXOMARK
I realize that DxOMark and DxO Labs are separate entities now. But it’s nevertheless suggestive of what DxO might consider to be important.
Given the very high resolution of the Z8 sensor, I wouldn’t worry too much about moire with it, unless your main subject happens to be moire-sensitive, such as bird plumage, fur, clothing with fine details or a noticeable weave, or patterned structures at a distance. If moire is only an occasional problem (as it is for me), you have options: use local adjustments to improve the rendering or use other software that demosaics differently, such as darktable or RawTherapee or the camera manufacturer’s own RAW development utility, and see if the result can be improved to your liking.
I checked a batch of images with moire on feathers with lens softness global = 0.
So here is what seems to be general behavior for my bird photography with D850 :
PL6XD vs PL8XD2s :
without moire correction :
- sharpness :
sharpness decreases more rapidly with XD2s as you move away from the sharpest parts of the image.
For me s (from XD2s) would mean softer. - pattern :
Due to sharp zone area smaller with XD2s, pattern is sometime less visible in less sharp areas, and sometime other larger pattern appears in these zones.
And some details are lost in these zones. - hue irisation :
Hue irisation is better attenuated with XD (but not perfectly) than with XD2s. - luma irisation :
when destaturating images it appears that “luma irisation” is way better attenuated with XD2s than with XD. - overall color irisation :
general impression is in favor to XD since XD2s less good hue correction overcome it’s way better luma correction. - overall impression :
overall irisation + more sharpness gives overall impression in favor to XD.
with moire correction :
-
XD moire correction :
_ Very little decrease of hue irisation.
_ Luma irisation does not change, but luma seems to change in not affected by moire zone, so would need further investigations.
_ Pattern does not change. -
XD2s moire correction :
_ More obviously decrease of hue irisation than XD.
_ Luma irisation : same as XD.
_ Pattern : same as XD.
Overall impression :
When luma irisation is intense, XD2s with moire correction is better than XD.
When luma irisation is not intense, XD2s with moire correction is worse than XD with or without moire correction.
Sharpness gain (related to lens softness) is better with XD.
Blurred backgrounds are better denoised with XD (often slight blotched pattern result with XD2s when XD is smooth).
Below are some image to stack (layers) in a pixel editor for instant comparison (be sure to download images at 100% size - not always the case) :
EDIT : I’ve just noticed that the forum resizes images above a certain size.
At the bottom of this post are more severe crops which should keep original size.
ZONES
DPL6 XD
DPL6 XD + Moire 100%
DPL8 XD2s
DPL8 XD2s + Moire 100 %
EDIT : I’ve just noticed that the forum resizes images above a certain size.
I’m resizing them to a tighter crop to avoid confusion.
DPL6 XD
DPL6 XD + Moire 100%
DPL8 XD2s
DPL8 XD2s + Moire 100 %
1 Like
A general question.
How do you judge moire. As far as I understood it’s becoming visible with zooming out. Yhe view examples I have don’t show moiree at 100% but start showing at a lower percentage.
George
What do you call this ?
DPXD2s + moire 100% - zoom 100%
