Thing is, the product of demosaic provides the starting point for anything other than capture sharpening, and the choice of algorithm has significant effect on the resolution of fine lines.
RawTherapee IMHO has put a lot more work into the demosaic and sharpening parts of the workflow than darktable. The converse is true regarding color and tone management, so one should choose their weapons with care and consideration…
Although I can see the extra sharpness in the RT screenshot, it has also introduced more grain into the sky and clouds. Even if it is a little softer, I would be happy with the DT result. I think we could all benefit from not obsessing over sharpness.
Edit: I got OP’s pictures back to front. There is more grain in the DT version.
Ah, you’re absolutely right! I mistakenly thought the top one of OP’s post was RT. So I take back what I said and admit that RT is a lot more impressive. So is RT applying its sharpening locally rather than globally through edge detection or something?
In RT you can set a contrast threshold to control sharpening. Based on that threshold only high frequency regions are sharpened. It’s the same algorithm which is used in dual demosaic in RT (and since a while also in dt dev builds dual demosaic)
Here is a quick try with the new “diffuse and share” module in dt. Considering how good is RT, I think that this module is promising. There are some edge artefacts but I am just discovering this module, thus lot of room for improvement ! Took color out to facilitate comparison.
Thanks! So this sounds very similar to the masking feature in Lightroom from my Adobe days. It was actually one of my favourite tools. You set your sharpening level, then hold down the Alt key while adjusting the slider to mask out areas of low frequency. Nice and quick. I suppose the Contrast Equalizer in darktable should be able to fulfill a similar role, right?
The edge tab does not directly act on edges in an image; rather it adjusts the level of the edge awareness feature of the wavelet transform. If you have not adjusted the luma or chroma splines, adjusting the edge awareness spline will have no effect.
Well, for example we could look into the code to see if the implementation of Amaze demosaic is the same in DT and RT or it’s different. That could confirm or exclude the demosaic algorithm being the root cause.
And if it’s not the demosaic, I see no other possible reason than the capture sharpening being better in RT.
Are you comparing exported jpegs or screenshot of the view in the corresponding tools? It could be an interpolation for the display, so I think the check should be done with exported jpegs.