Open teplit opened 4 years ago
Well, for a better comparison you should set the amount in RL to 100. Your setting of 50 simply blends the unsharpened with the sharpened one at 50% which obviously gives a less sharp image while suppressing the halos in RL.
Left is Capture sharpening, right your RL settings but amount at 100 to show the plain RL
RL in detail tab has a lot of bright halos.
Maybe I should add an amount slider to allow blending unsharpened with capture sharpened the same way as RL in detail tab does ;-)
another one, same crop as last one but this time right with RL amount in details tab set to 50 as in your settings. Still bright (or white) halos
For even better technical comparison between Sharpening – RL Deconvolution and Capture Sharpening all relevant settings of both tools need to be equal. In addition to that for fair comparison the style for RAW development needs to be the same – a darker image emphasizes white halos.
So here is an image with Sharpening – RL Deconvolution only:
And here is such with Capture Sharpening only:
As a whole the last image looks uglier to me.
There is a reason, auto limit iterations is enabled by default in Capture Sharpening. I would never use 100 iterations and I always use the auto limit iterations enabled except for this kind of technical comparisons.
Btw: in my example screenshots above I did not use 100 (I used 30) iterations for RL in details tab and the bright halos were still present
Here's another comparison. Middle is unsharpened. Which one has more halos, right or left?
I noticed the bright halos in the right image, but the problem is that Capture Sharpening causes more dark halos with the same settings as Sharpening – RL Deconvolution. That is evident from the previous comparison where the "brake" of Capture Sharpening is switched off.
There are cases where even 100 iterations of RL Deconvolution are not enough to obtain sharp looking image with acceptable amount of fine detail (increasing the radius beyond 0.7-0.8 is not a solution for me). Unfortunately I cannot provide a RAW file to exemplify this problem, but for some problematic photos I do get sharper results with more fine details from Camera Raw. The given RAW file is a sample produced with expensive equipment and does not need a lot of sharpening anyway.
but the problem is that Capture Sharpening causes more dark halos with the same settings as Sharpening – RL Deconvolution:
Trust the auto limit iterations to reduce the dark halos. The result will not be perfect (it can not be perfect as we do not know the exact psf), but imho it is much better than the other RL deconvolution in RT. It works on different data (CS works on linear data, while Sharpening RL deconvolution works on Lab L channel (non-linear))
Unfortunately the documentation of CS is not in rawpedia. But generally it is better to use CS with rcd-demosaic instead of amaze-demosaic because amaze-demosaic is more prone to generate dark artifacts than rcd-demosaic and CS then enhances this even more.
There are cases where even 100 iterations of RL Deconvolution are not enough to obtain sharp looking image with acceptable amount of fine detail (increasing the radius beyond 0.7-0.8 is not a solution for me)
Neither of both (using such a large amount of iterations or increasing the radius beyond what is correct) is what I would do. I usually stick with the default of 20 iterations (limited to even less iterations by auto limit iterations) and (rarely) lowering the radius.
If 20 iterations are not enough to give you more detail then most likely the radius is wrong. Though I gave my best to auto-calculate the radius it might be still sometimes not perfect...
Unfortunately I cannot provide a RAW file to exemplify this problem, but for some problematic photos I do get sharper results with more fine details from Camera Raw.
That's sad, that you can not provide a RAW file :frowning:
I noticed the bright halos in the right image
They are not problematic for you? I find them much more problematic than dark halos as bright halos are more eye-catching than dark halos.
Of course the bright halos are problematic - that's why I used initially 50 for the amount of deconvolution.
I decided to share conditionally a problematic RAW file in order to exemplify the difference between Capture Sharpening and Camera Raw sharpening.
First is the RawTherapee's automatic Capture Sharpening (with RCD demosaicing):
Second is Camera Raw sharpening:
The RAW file is available here for one week: https://filebin.net/8u1poslf6z2g7964 I give you permission to use it only for technical testing of open source projects.
The Camera Raw image obviously contains more artifacts, but it is less fuzzy. When I view this image 1:1 on 24" 4K monitor (~185 pixels per inch) the artifacts are OK for me (yes the sharpening is bit high, but I believe that with less noisy camera sensor the same sharpening will look better).
Thanks for the providing the raw :+1:
I agree that for this image the default of 20 iterations is too low
Capture sharpening default (20 iterations)
100 iterations
This photo was taken several years ago with higher aperture then necessary (f/16) and that most likely caused considerable softness. Much later I found that the maximal sharpness for these lens is around f/8. Nowadays I take into consideration this, but sometimes I still get surprisingly soft photos – the problem is that I don’t see clearly the sharpness of the image on the display of my mirrorless camera.
In my opinion the RL Deconvolution of RawTherapee do give better results than Camera Raw’ sharpening for relatively sharp photos (for which 0.7 radius is enough).
In my opinion the RL Deconvolution in Details tab of RawTherapee (btw: Capture Sharpening also is RL-Deconvolution) can look sharper (though it is not) than Capture Sharpening because it has more halos than Capture Sharpening. Capture Sharpening tries to avoid this halos.
Another advantages of Capture Sharpening are the Corner radius boost (which increases the radius of the gaussian deconvolution linearly depending on the distance to the center of the image) and the automatic settings (finding good base settings for RL deconvolution in details tab is very time consuming, while the auto radius calculation at least gives a quite good starting value)
A screenshot without any sharpening:
A screenshot with RL Deconvolution sharpening only:
A screenshot with automatic Capture Sharpening only:
The RAW file is located here: https://cdn.hasselblad.com/samples/x1d-II-50c/x1d-II-sample-09.3FR The content of the PP3 file:
I don't see a practical reason to use the new Capture Sharpening tool.