[Updated 1. title] Disappointing result some cases (high ISO) with PL8 vs PL7 on Lens softness correction?

andras.csore · September 17, 2024, 9:55pm

[Original title]: Disappointing result with PL8 vs PL7 on DeepPRIME XD/XD2s

[Update 1]: In the meantime, i think its not related with XD2 vs XD2s (seems XD2s is usually better), but with ‘Lens softness compensation’…

[Original]:
I check thru in 7 photos (my own), how PL8 is better on that than PL7. And approx 4 from 7 cases PL7 results of far better, others may equal. PL7 results more detailed, and even background is more smooth. I not talk about very-very small differences. Pretty visible in 1:1 (100%)

Other results / Experiences on that so far?
I check a few review on Youtube: Peter Forsgård ; Matti Sulanto, but seems they show like one image, with good lights, etc, and PL8 look a bit better.

May its in my raw files (may this image breaks the logic), my camera, lens, PC (Win11) or GPU config or some constellation of the things. I don’t know. But difference is visible at first sight. I use the same .dop. Exported and compared to tiff files. May i miss something.

Example: Left is PL7, right is PL8. I pick up the ‘most visible difference’ image/section. Zoom: 100%

Please, can someone double check? RAW file and .dop attached below.
Body: Olympus EM5.Mk3.

Raw file:
20240831_210554_521 Copy.ORF (20.8 MB)

Edit: for some reason i upload under wrong filename (copy of). But i not able to update/replace… Pardon. But that’s the raw.

orf:
20240831_210554_521.ORF.dop (14.1 KB)

Additional question: How i know in PL8, its use XD or XD2s during export?

Andras, long time DxO user…

platypus · September 17, 2024, 11:16pm

Exported the images as 16 bit TIFF and compared them in Lightroom.
Differences exist, but I find them to be negligible.

I also noticed that you reduced denoising to 33/40 so I created virtual copies and exported those at 40/40. Differences still exist, but I find them to be negligible again.

John-M · September 18, 2024, 6:04am

It’s always been the case that the difference between DeepPRIME and DeepPRIME XD~ has been mainly evident when an image is heavily cropped - in which case, the XD version is better … where the degree of “better” can depend on attributes of the particular image.

My approach is to use DeepPRIME (standard) by default – check the results – and use DP XD~ only if I’m wanting a “better” result.

zkarj · September 18, 2024, 6:39am

I cannot disagree more!

XD2s will be my default because it routinely beats DeepPRIME XD on my challenging images as it does not have the side effects of XD (invented noise in smooth bokeh).

For the run of PL7 I have defaulted to DeepPRIME because of this issue. No longer.

andras.csore · September 18, 2024, 8:58am

Many thanks! Seems its may depend on software/hardware constellation, or something like that. Regarding denoising 33/40 - i prefer 33, still quite good results, bit more details (as at least i think its may a sweet point for Oly images, than 40).

andras.csore · September 18, 2024, 9:00am

Ok. So, you get better results. Great! Sounds good.
May if i ask, can you test PL7 vs PL8 with my raw file? (see in the post).

Fineus · September 18, 2024, 9:01am

Each to their own, but I rarely use XD and may not use XD2 as I found it introduced artefacts into my images that I wasn’t happy with, while regular DeepPrime did a great job.

platypus · September 18, 2024, 5:05pm

Retaining some noise helps to make the image look sharper too…

I had experimented with settings lower than 40 and found that the reduction in noise reduction (!) was not progressing linearly. Occasionally, I add some “grain” back to a fully denoised image to make it look more natural. Oldtimers somehow prefer good old film grain to fully denoised (plasticky looking) images

Note: Film grain comes with FilmPack

zkarj · September 20, 2024, 4:34am

No, I was speaking from my experience testing with my own images, including ones that XD struggled with. Without exception, anything ISO 25600 or lower looked fantastic with the default level of XD2s.

I did do some testing of low ISO images, and found no issues I could notice. For me, where it really matters is the high ISO stuff which is almost unusable without effective noise reduction. DeepPRIME has been my default (on all images). DeepPRIME XD2s will take its place with PL8.

BHAYT · September 20, 2024, 1:28pm

@andras.csore My current camera is a Lumix G9, a similar product to yours, and I do not consider DOP XD2 or XD2s to be automatically superior and for your images I get what I saw when testing the product, namely that the new product typically produces softer exports than the old product i.e. PL8 left and PL7 right and exported as 100% JPG

@200% we have

I also created a VC so that VC[1] contains your image and edits and [M]aster contains the edits without any NR and I compared those [M] exported images using FastRawViewer with the sharpness viewers and got

and

showing that the images before applying NR are somewhat different, PL8 left and PL7 right ?

But using the same technique with the NR exports simply doesn’t really help but seems to show greater similarity than difference?

Does that help confirm what happened on your system?

PS:- The red sharpness feature of FRV frequently picks up noise as well as “sharpness” points or edges.

andras.csore · September 22, 2024, 9:17pm

Many thanks for your tests and descriptions, especially for the ‘Sharpness’ alike comparison! That was the point when i get the idea.

In the meantime i continue to find out the thing (not finished yet). But as preliminary, i can say:

XD2 vs. XD2s usually near the same. XD2s tends to be a bit cleaner, usually background and background highlights better, ‘dithering’ parts definitely better (by a bit). Sometimes may a bit softer, but negligible and rare cases (if any, hard to tell, may i not see the forest from the tree…).
So i think at the moment, XD2s is same or better than XD2. Definitely not worst. Also a bit faster (like 10-20%?) to render.
XD2s definitely provide less artifact than XD2 (at least what i recognize, like 1 from 10 photo definitely less artifact)

And in the meantime, when i start to turn on/off the ‘Lens softness compensation’ → seems (at the moment), that where thing may going to different PL7 vs PL8.
In ‘good’ ISO200 portraits, the result is near the same (or near nothing difference). But in less optimal light and for example ISO5000 → PL7 seems more sharper (zoom: 100%), or PL7 more ‘aggressive’. To be more exact: seems in PL7 larger areas sharper (and it think, that’s the point). As in most of the times, sharpness in part of the image is the same or similar (like some part of the hair), but in the skin part PL8 definitely softer OR PL7 was ‘over sharped’ a bit.

Note: new features in PL8 : “Upgraded lens softness compensation, free of fringing and artifacts”

I continue the journey, i let you and the forum to know more.

One other example in the meantime: PL7 left, PL8 right

1.) Less sharper
2.) Sharpness is okay (less aggressive)
3.) May too smooth
4.) Background is better
5.) Edges in the ear is better
6.) Texture of the shirt is more natural (less aggressive)
7.) May too smooth

BHAYT · September 24, 2024, 10:00am

@andras.csore Please post the image and DOP and then I can scrutinize the exports but I still see the PL8 export overall as softer!?

With ‘Lens Softness’ my normal setting is -0.5 and I have yet to test that and variations with PL7 versus PL8, i.e. with& DP XD versus DP XD2s.

In the meantime here is a test from a recent trip to Brighton Royal Pavilion at ISO 25600 taken with my G9 (only the Mark 1).

PL8 (Left) PL7 (right)

P1137215.RW2 (23.0 MB)

P1137215.RW2.dop (48.8 KB)

The DOP is a PL7 DOP which can be used for both PL7 and PL8. My strategy when testing is to keep the images etc. separate to avoid PL8 changing the DOP which PL7 will not then be able to “read”.

The DOP tests are

XD(2s) at default
at Force Details (FD) +50
at FD +100
at FD -50
DP , FD -100 is so “messy” that it is not worth investigating that I decided to see what DP made of the image!

where item 0 is the [M]aster image, 1. is VC[1] etc.

I have another test on a different image (the Egypt image from the Forum Benchmark spreadsheet) with FD settings from 0 to 100 in steps of 10 in one directory and from 0 to -100 in another directory and I should apply those DOPs to the image included here.

uncoy · September 24, 2024, 11:27am

DeepPrime XD2 finally looks usable. DeepPrime XD was a train wreck with artificial detail everywhere in real world. In low light environments, edged of faces would be covered in crawling maggots.

DeepPrime was brilliant and didn’t have these issues. I’d prefer to see comparison between DeepPrime and DeepPrime XD2, or at least the original DeepPrime included in the mix. Perhaps DeepPrime XD2 is more or less rebranded DeepPrime?

mwsilvers · September 24, 2024, 12:19pm

Zoomed out the differences may not always seem to be very great, depending on the image. If you zoom in a bit, the differences between DeepPRIME and DeepPRIME XD2s are very obvious.

XD2s may be more useful on cropped images. Compared to XD, at least on the images I have applied it to at this point, It provides more natural looking fine detail with far fewer artifacts than XD. Like in all things related to quality, my opinion is somewhat subjective. Others may not agree.

Mark

JoPoV · September 24, 2024, 12:45pm

I’m fighting with that now (here pl6 and xd vs pl8 and xd2s) :
3 constants I’ve found :

v8 softness default values gives softer result than v7 softness default value.
when increasing v8 value for a “better” comparison, i find that v8 sharpness gain decreases faster than v7 in soft focus zone.
if compare to classical unsharp mask terminology, it is like radius seems bigger in v8 than v7 (maybe this depends of where in "zone of sharpness).

It could be expressed as follows : sharpness gain is concentrated in a more little dof zone (or sharpness zone) and default value gives softer result in v8.

BHAYT · September 24, 2024, 1:01pm

@uncoy The image and DOP that I provided had DP applied to VC[4] and if you take my image and apply the DOPs you will get this for 0 ([M]) versus [4] (please remember this was an image taken in very low light to protect the colours of the artefacts another use for the word “artefacts”).

So we have

PL8 - XD2 versus DP
PL7 - XD versus DP

I have labelled the first!

There was a problem with the PL8 DP export, I left the FD at -100 so I have updated the snapshots and all should be O.K. I hope.

The revised DOP

P1137215.RW2.dop (48.8 KB)

uncoy · September 24, 2024, 1:33pm

Thanks for the clarification. It’s important to get the lens sharpening at an equivalent level before starting the noise reduction.

Using default noise reduction setting (40) is almost a waste of time. There’s no detail left at that value. I’m usually working from 0 to 12. The noisiest images I’ll move up to 20. Of course the value varies depending on which version of noise reduction one is using (HQ, Prime, DeepPrime, DeepPrimeXD). DeepPrime requires very low levels for a strong effect. I like to leave a bit of natural grain in my photos. YMMV.

Thanks for the tests in any case.