Hi Giuseppe,

Nothing has changed in Markesteijn demosaic since 2.4, so far as I know.
The algorithm is "upstream", as it comes from dcraw. You could try
Frequency Domain Chroma (which is code written for darktable) and see if
you're finding better results. There aren't many cases where VNG will be
preferable to Markesteijn, either 1-pass or 3-pass.
You're presumably seeing maze pattern artifacts. (You can also sometimes
see them in the EVF of a camera with an X-Trans matrix when it is in
zoom mode -- this isn't exclusive to darktable.)
From
https://eng.aurelienpierre.com/2018/11/30/filmic-darktable-and-the-quest-of-the-hdr-tone-mapping/#known_issues
it appears that filmic doesn't work well with noisy dark tones in X-
Trans files without manual tuning.
I don't have a recipe to get rid of these artifacts (others may). The
good news is that something which looks pretty rough at 100% may still
look fine when printed (or on a screen).
Dan



On Fri, Dec 21, 2018, at 6:54 AM, Giuseppe Pagnoni wrote:
> Hello,
> 
> I am using the latest release version of DT 2.4.4 (which is great and
> a big thanks to the developers!) and a camera with an x-trans III
> sensor (Fuji X100F).  While everybody's recommending the markesteijn 3
> pass algorithm for demosaicing X-Trans sensors, I have noticed that in
> dark areas of high ISO images this algorithm produces an unpleasant
> artifact, namely it tends to render noise in a kind of "tiny sticks"
> pattern, instead of the more organic original dot-like grain (you have
> to zoom in at 100% to see it).  This artifact is not produced with the
> VNG algorithm, which on the other hand seems to create other problems
> (e.g. near high contrast edges) that are better managed with the
> markesteijn algorithm.> 
> Any thoughts on this and how to solve the issue?
> 
> thanks in advance for any suggestions
> 
> PS: I have tried the 2.6 RC2, which is impressive (e.g. the defaults
>     for noise reductions are *much* more sensible, and the filmic
>     module is awesom).  This particular issue, however, seems to have
>     remained unchanged> 
> giuseppe
> 
> 
> 
> _____________________________________________________________________-
> _______ darktable user mailing list to unsubscribe send a mail to darktable-
> [email protected]

____________________________________________________________________________
darktable user mailing list
to unsubscribe send a mail to [email protected]

Reply via email to