Showing results of different crops/magnifications/(maybe even lens corrections) is confusing.
Also the OP RT version had more than CS enabled (usm sharpening, wavelets), while the result of my last post had only CS enabled
1 Like
In my experience for the majority of images a 5x5 kernel (which is used in RT up to sigma 0.84) is enough.
There are rare cases (for example shots at >= F22) where a sigma > 0.84 is needed.
I was agreeing with you on the CS for sure and I think it is just his screen shot because when I load his PP3 it looks way less exposedâŚmore like yours so it could just be a web thing
OP image with their sidecar
This doesnât match to me with what they posted originallyâŚmaybe just my eyesâŚ
To be honest I didnât try the pp3 sidecar of OP, I only looked at its settings using a text editor 
That doesnt look like RT Is that your vulkan project??
yes, but itâs still kinda ingoâs code. i just ported it to glsl.
The sigma I used was the one from automatic calculation, for which I still canât explain why it works, you know 
But Iâm still working on it 
For sure just noticed the interface in your screen shotâŚ
Hereâs a comparison. All from RT
Top left: no sharpening at all
Bottom left: RT Capture Sharpening
Top Right: RT Capture Sharpening + a small amount of USM
Bottom Right: RT Capture Sharpening + a small amount of USM + Microcontrast
Edit: for the bottom right (with microcontrast enabled) I would like to mention that in the RT pipeline Microcontrast is applied before USM. So USM enhances the effect of Microcontrast. Would be worth a try to change the order of both in RT pipelineâŚ
2 Likes
Have DT devs ever spoken about porting RTâs capture sharpening? It can make such an amazing difference.
1 Like
1 Like
Good and funny example: that cherub (or whatever it is) is definitely wincing at the glare of the sun now!
1 Like
I think there might be a Lua script??
We donât need it, I have better.
Diffuse and sharpen module. Using second order partial derivative equations of anisotropic heat transfer in wavelets space. Also works for dehazing or increasing local contrast. Has 2 different ways of avoiding edges to prevent halos.
Itâs kind of slow, though.
The same algo can do:
-
surface blur (just with different parameters):
-
Blooming:
-
Dehazing
-
Inpainting (although just on small parts and I donât reproduce the results of the litterature).
5 Likes
Certainly looks capable of some heavy lifting. Love that bloom too.
More subtle sharpening (similar to the RT results in the first post) is also possible! IMO @anon41087856âs is a little overdone.
1 Like
One thing to take into account is I used the OP XMP, which overdoes the black thresholding. It still looks good zoomed at full-picture level, and I would advise against pixel peeping to setup this kind of algo (unless you need to assess the technical perf of the algo â but thatâs not retouching anymore), since nobody goes into museums with a magnifying glass to count the number of threads of the canvasâŚ
1 Like
Another version using diffuse/sharpen but also using the details masking (which is available in darktable 3.6) to exclude the sky. Still not perfect for the pixel-peepers but Iâm still getting my head around this module:
Either way, this module is magic, especially now it has OpenCL support.
1 Like
Agreed
The dark horizontal line between sunlight and shadow towards the bottom of the image is easy to catch without pixel peeping, though without having used the module, Iâd assume that can be avoided. Most instagrammers probably wouldnât notice it, but this is pixls