I see Color balance RGB as the replacement for shadows and highlights, not tone equalizer. I only use the latter if I can’t get what I want with the former.
Just saying, those sharpen settings are still subjective decisions, made by someone else on your behalf, for some theoretical “average” image.
This is way too much sharpening IMO, which comes from the bad habit to view 20+mp images at 100% or more. We have all been there.
A more sensible approach to sharpening is to split it in three phases:
- capture sharpening: only correct the effect of the antialiasing (AA) filter of your camera sensor. Note that many recent cameras don’t have this anymore, so you can skip this step
- creative sharpening: make the image look crisp; this is mostly fake sharpening, a.k.a local contrast
- output sharpening: counteract blur due to resize or prepare the image for printing. DT does not handle this natively, but there are lua extensions
See this example. My camera, DPR studio image with a very sharp lens
see the halos around high contrast edges
For example, see instead a more gentle approach to capture sharpening
with these settings
It was never intended to be used withut further adjustment as needed
I wanted the preset to be able to over-sharpen and undersharpen. I set it initially so it worked on many of the types of images I shoot, and I am fine tuning it as I see the results on more images. But even then, I always intend to adjust the amount of sharpening after the initial values are applied. If I had worked on your test image, I would have reduced the sharpening that the initial settings are at.
Why limit fake sharpening to local contrast methods?
What do you mean by this? I do not set sharpening by pixel peeping, if that’s what you mean by 100%.
With “local contrast” I don’t necessarily mean the local contrast module in DT, but all those “sharpening” methods which actually only increase acutance (for example contrast equalizer).
In this recent video AP explains the difference between resolution and acutance, and why our eyes are not very good at separating those. See 9:15 for example.
What I want to say is that any recent high MP camera + decent lens is plenty sharp already, to an extent that we old timers, grown in the analog film era, never experienced before.
Digital imagery brought the possibility to see pictures a 100%, which is equivalent to look at 1m print from few cm. It can be done, but it is not the normal viewing condition.
And at 100% any camera + lens combo, no matter how good and expensive, looks unsharp.
Therefore the overshapening habit that we see everywhere.
I am talking in general here, not about your specific case.
Cool. I noticed in that video that he is talking about using the edge sensitivity control the way I use it (I noticed your example did not use it at all, and seemed to rely on minimising other settings to avoid edge artifacts)
Yes, I agree and this is what I meant by pixel-peeping. I do not do this when sharpening. Just about the only time I might get in close is to check for noise, and even then I don’t bother if it already looks good.
On the other hand and not to disagree with anyone wrt sharpening but I have found the only accurate preview for an export is what you see at 100%… the DT preview as you zoom out scales and at least for me hides noise and detail pretty quickly…so you can call it pixel peeping but you do have to experiment a bit to get to know DT. There was an old thread … ( Denoising high-ISO images: a sourdough bread - #54 by Underexposed ) The OP wanted to know why his image had so much noise still after denoising when he saw none on the screen. It was a black and white version and really you only saw it at 100% and what you saw there matched the export… there can be other artifacts as the one I noted just the other day playing with a playraw image… Windows build of R&Darktable.... - #56 by priort So what I am saying is there is also the factor of what are your preview conditions and are they a reliable representation of the final output and as @MarcoNex mentions the intended media and viewing conditions for output will play an important role as well… The above example is related to noise but would likely apply to sharpening as well…
Try to increase the speeds and/or iterations.
I tuned this preset for my camera with a studio image and a sharp prime lens.
Real world situations with consumer zoom lens or small motion blur/misfocus, might require a more aggressive sharpening.
Yes, that’s how I formulated the preset I was using - by increasing the speeds and other settings. I found I didn’t need extra iterations - as you said as well, it is capable of over-sharpening anyway - even in one iteration. So maybe I could use smaller speeds and more iterations, but that would make it slower wouldn’t it?
It was then I discovered how the edge awareness was a good tool for controlling artifacts (halo). Settings like radius span and edge threshold were like the fine tuning for sharpening - and I suspect will change, particularly with sensor resolution.
Hi @MarcoNex . I was wondering if the image is available for download to test out myself. Thanks.
You can get one from dpreview or imaging resources.com… Likely they have one shot with your camera…
EDIT So for example here is a base iso for the XT-4
DSCF2543.RAF (56.0 MB)
There is haloing to be seen around edges and letters, even on a downscaled version. Doe me this is overdone and not something I would like. But that’s me.
edit I would still really like a raw file from you, to toy with. It’s perfectly fine if you don’t want to, of course.
On which one, there’s four I have posted now
How would I give you a RAW. It’s 21 MB - is that too big to attach here?
Raws are uploaded all the time…
P1050979.RW2 (20.6 MB)
You probably need to add some sort of license, like a Creative Commons indicator or something else.
Since there are no noise-profiles for this camera, it’s more of a balancing act to get the denoising right. I think this helps a lot in how much sharpening is needed.
If I do this as I usually do images, I ended up with a default instance of ‘local contrast’ at the end (nothing changed, so the laplacian mode at default settings). In front of that (so under it) is a ‘local contrast’ in bilateral mode, with the contrast set to 3, and then adjusting the ‘details’ slider to balance the noise. I couldn’t really push it high.
As for diffuse, I enabled the default ‘demosaic: AA’ preset, and moved it under (so before) ‘input profile’ (I guess that’s still the correct spot for this?). I added another instance with the ‘lens deblur: hard’ preset, and kept it at the default spot, which is just above ‘color calibration’, but before/under ‘filmic’.
Exported at full resolution, then resized down to some output settings with a bit of sharpening applied to the final output, before saving as JPEG.
So these JPEGs are meant to be viewed at 100%.
P1050979.RW2.xmp (25.0 KB)
(PS: I added a bit more sharpening after the resizing, as I normally would. I guess I’m trying to compensate to your sharpening level )
Compared to the preview-JPEG that’s inside the RAW file, I think (as in, in my opinion) this is sharpened just fine.
The example you gave in Still need 'sharpen' module after D&S ? - #34 by n01r has artifacts of trying to over sharpen things (haloing around the ‘RUSH’ text, the sailor itself (particularly the legs) and the left side of the sail (the blue part against the sky)).
The shot is just quite blurry. It’s a compact camera that is trying to do a 500mm equiv zoom on a somewhat-moving subject. So, I’m guessing you shot wide open at f5.6, at a quite zoomed range on a lens that’s not known to be sharp.
Also, it’s known that this model applies over sharpening to its JPEG files.
Now, since that might come over as negative, let me sort of rephrase it:
You tried to take a difficult shot with a bridge-camera, and the result is probably at the limit of your gear, but not as good as you hoped. So you try what you can to get the most out of the shot. Nothing wrong with that.
But that means you’re trying to get more sharpness out of the file than your camera / lens recorded, so you are over sharpening in this occasion, because you ‘have’ to. So yes, you apply more sharpening than - for example - I would in my ‘normal’ shots.
But I also had shots like this, where I wish the result was better, but I have to make due. Do what you can in that case :).
(PS: The ‘contrast equalizer’ also has different ‘deblur’ presets. The large-radius strength 3 preset might be a replacement for the older ‘sharpen’ module in this case?)
An easy way to get ‘more’ out of diffuse quickly, is for me to still use the ‘demosaic AA’ preset, but increase the iterations. It quickly goes into ‘too much’ territory. In your case, setting it to 3 is already too much, IMHO. But then you can use the uniform-mask to change the opacity of the effect, to bring it back in line where you would want it.
I thought I’d load the file into DxO Photolab, to see what it makes of it,.
Now, the default, ‘profiled’ sharpening that is applied (‘lens sharpness’) in DxO is supposedly being adjusted for camera/sensor and the lens and the settings at which you took it.
Normally, I can’t hardly tell the difference between DxO’s sharpening on and off, on my fullframe Sony A7m2. But if I load files from my Olympus m4/3 camera, there is a clear difference. So DxO thinks my Olympus should be sharpened more than my Sony.
Your file, is very much over sharpened in DxO. They think ‘man, this needs so much’ that they overdo it .
(Clean DNG export, with only corrections and the ‘micro contrast’ applied:
P1050979_fix.dng (68.8 MB))
No, I didn’t take it as negative. I could have used shots from my Canon 80D with a Sigma 150-600mm lens, but I tried to pick a difficult example.
Since posting my own D&S preset, I have a new variation with much less on the speed sliders and I use it at 2 iterations. 3 is too harsh - so thanks for the idea on using the opacity mask.
Also thanks for your time and effort. Among the things you detail are things I need to experiment with.
That is interesting. I also recall you observing that the FZ80 is known to “oversharpen” it’s own jpegs. Good to see what DxO did, that’s for sure.