RT vs DT - sharpening - 1001

Tamas_Papp · March 27, 2023, 11:59am

Yes, I am basically in the same boat. I wish I had time to dig into the math, the connection between speed and anisotropy for various orders is not something I understand at the moment. Cf

github.com/darktable-org/dtdocs

diffuse or sharpen: document PDE and its relation to speed and directionality parameters

opened 01:31PM - 14 Feb 22 UTC

tpapp

no-issue-activity

It is somewhat difficult to understand how the 1st, 2nd, 3rd and 4th order speed… and directionality settings should be interpreted for [diffuse or sharpen](https://github.com/darktable-org/dtdocs/blob/9c07a8077d102a7121409d8a5fdc869d428f380a/content/module-reference/processing-modules/diffuse.md). Specifically - the manual talks about high and low frequencies, it is unclear if this is a consequence of the implementation or a property of the PDE, - gradient and laplacian are used, but the gradient is a vector while the laplacian is a scalar, it is unclear how they mix I would suggest 1. Documenting the actual PDE that the module *intends* to solve, abstracting from the implementation details. While not all users are familiar with PDEs, it would help some people and more importantly they could contribute explanations or examples based on this. 2. Document the implementation details to the extent they are relevant. 3. Clarify where frequencies enter this picture: as an implementation detail or something fundamental to the PDE.

lphilpot · March 27, 2023, 12:15pm

I’m even less… I don’t have the math background to even understand it.

priort · March 27, 2023, 12:19pm

@nwinspeare Nicolas did a really nice job and he had a nice example image where he demonstrated the tweak on image detail… Having said that as I mentioned earlier 95% of the time is use a modified preset… I find the no AA preset does a good job and then you can dial up slowly or or two or maybe at most 3 iterations in IMO… I know everyone’s opinion of good sharpness varies… You also really have to check the result at 100% I find the DT preview is often very misleading for noise and sharpening … it might look fine full screen but if you zoom its cooked or you have introduced a lot of grain that just seems to get lost when the image is scaled to give the full screen preview.

clind · March 27, 2023, 12:39pm

Thus very one thread, thanks
Hope it’s still an interesting read and not too outdated

jonathanBieler · March 31, 2023, 6:25pm

I’m testing the last DeepPRIME XD model from DXO which has lens deblurring option, and somehow it’s less sharp than the original file I produced with RT. Turns out it’s because RT can’t apply capture sharpening to the DNG DXO produces. Capture sharpening’s pretty good.