Settings for fine details (compared to other tools)

The zoom rate is exactly 100%, so I would think, it shouldn’t interpolate whatsoever - I, mean … 100% is the level you want to be as accurate as possible, to check your picture.

But in any case, the export doesnt’ look different. I checked that. Here is a comparison of the exported JPG

2 Likes

Can’t see much difference in the first pic either. In the second picture I’m unsure… might see a bit difference in the scratches, in the text and the metal thingy on the top right … these all seem a bit clearer in ART.
But yeah, it’s not as defined as in other examples.

2 Likes

You’re right, it’s definitely there in the export as well, intriguing… Specially since ART looks very natural and it’s not trivial to achieve a similar effect in DT

2 Likes

I thought it might have to do with the raw format, as the best samples here are .ARW (Sony) files :man_shrugging:

Need to check that.

2 Likes

With this one, I also see ART as sharper.

Nope. Here is an example from a .cr3 file, also a play raw…
(https://discuss.pixls.us/t/the-leprechauns-must-be-lost)

Gonna let the dust settle a bit :cowboy_hat_face:

In these examples, what diffuse or sharpen settings are you using? :smiley:

The detail has to be there in the scene and file. @123sg examples above for instance (rodeo and car) have very little fine detail or texture.

I’m an RT user and the detail you get from RT/ART has been unrivaled for years. At least in practical terms as you can probably get similar results with other software but with RT/ART it’s just there.

It’s worth noting that capture sharpening in RT/ART does give a bit of pixel level “grain” to blurred transitions. Wide continous transitions become a bit noisy so for blokeh fest photos it’s worth tuning down or off.

3 Likes

I’ve realised that now… :grimacing: :laughing:
Been playing with the last image @qmpel used as example, fiddling for the last 20 minutes with DorS, trying to replicate that clean yet pin sharp result from ART.

Can’t… (yet?!) this is as close as I got.

or maybe this…

Another potential downside is that this kind of thing is resource heavy - fine with reasonable GPU but very slow otherwise. I don’t know how demanding the capture sharpening in ART is though.

2 Likes

That looks really good already :+1: Gonna compare it in detail.

Now I request your settings :smiley:

Sorry, I see they are included in the shot

1 Like

Well, I see your settings are getting the image sharp at very fine level (I was also playing with a small radius and only the sliders 3 and 4, so we were thinking in the same direction, I think). But! There is a lot of noise and artifacts introduced, look into this 400% magnification of your screenshot and the ART reference.

[removed screenshot]

I think we are on the right track. Maybe tune it a bit down. I will try also with the edge sensitivity and sharpen sliders.

edit: faulty method on my side, showing the JPG enlarged obviously shows all the JPG artefacts, sorry 'bout that.

Here is the valid 400% comparison

This looks actually much better - if we tune it a bit down?!

1 Like

Not very demanding. No issue having it always on despite old hardware and the sharpening not being GPU accelerated.

1 Like

Are you removing chromatic aberrations? Looks like they are making things worse in the dt version. No sign of it in the ART version.
edit: also worth ensuring the demosaicing is the same.

I made an error while comparing, sorry. Edited the post above.

And here is a version where I replaced your darktable screenshot with mine (xmp file below). In my opinion it’s all a matter of settings. For dt especially DoS.


1X1A7276.CR3.xmp (8.4 KB)

Because little changes in settings can result in big differences, comparisons without posting the sidecars are not really usefull. Also pixel peeping at mor the 100% is a bit odd.

And anyway, “sharpness is a bourgeois concept” according to Henri Cartier-Bresson :wink:.

Certainly lots of detail but it has those points/pixels that are over sharpened. This used to happen a lot with RT and RL deconvolution sharpening.

Hope no one minds but here’s RT with default capture sharpening settings. RT automatically adjusts sharpening depending on the image. For this one it deemed an appropriate “contrast threshold” to be 7. As you can see the windows of the University building (just right of the copper clad church spire show similar detail to the dt file but with no “dotting”


I have to emphasize that this is without any intervention. Just out of the box with RT.

But we’re sharpening out demosaicing artefacts… The fenestration is basically vertical.
2024-01-09-123431_403x437_scrot

1 Like

My darktable 100% screenshots pasted next to your ART versions from the ART - dt comparisons posted above. Don’t do the comparison in the browser – at least with Firefox, the the images get fit to the screen, and Firefox’s upscaling does something weird. Download these a and compare them in an image viewer, or in Gimp etc.


MJD04337.ARW.xmp (16.1 KB)

I think you had substantially higher contrast in the ART image than in darktable, which can influence perceived sharpness.


1X1A7276.CR3.xmp (17.1 KB)

1 Like

You really managed to get the detail on on the bee one. The Zurich view looks “crunchy” and full of colour artefacts in comparison to the rather natural looking but more detailed ART image.

Depending on print and size the dt Zurich view might look better but with RT/ART you can add macro scale sharpening when needed. The initial sharpening just brings you to the maximum fine detail.

1 Like

He would know coming from a textile factory and land owning family!

1 Like

I think much of the colour artefacts are due to CA. I’ve tried to suppress them, and tweak the sharpening a bit.


1X1A7276.CR3.xmp (15.7 KB)