I guess that should go to the list, too. ---------- Weitergeleitete Nachricht ----------
Betreff: Re: [Darktable-users] Simple staggered demosaic code in dt works better than ppg or amaze. High iso demosaic noise issues? Datum: Mittwoch, 19. Februar 2014, 08:19:51 Von: Gert van der Plas - Schrama <[email protected]> An: Tobias Ellinghaus <[email protected]> Sorry I dropped the s. It's not nni. It is a weighted linear interpolation between nearest neighbours at the corners of the photosites. For r,g,b between resp 3,2,3 neighbours. Still very crude, but way better than nni :-) Op 18 feb. 2014 12:45 schreef "Tobias Ellinghaus" <[email protected]>: > > Am Dienstag, 18. Februar 2014, 12:20:50 schrieb Gert van der Plas - Schrama: > > Hi, > > Hi. > > > I've noticed that ppg and amaze demosaic create a kind of structured > > patterns. The patterns very pronounced with high iso images ceating color > > blotches and denoising profile on color causes unpleasant green spikes in > > the image. > > > > I've coded a simple staggered nearest neighbour interpolation demosaic and > > it seems to perform better than dt 1.5.2 I checked out of git. Less > > artifacts/patterns, color noice is more fine grained and profiled denoising > > works better. Sharpness is of course less but for high iso this is very > > reasonable. Here is a dropbox link to a 200% screenshot in my patched > > darktable-master. > > > > https://www.dropbox.com/s/l2vquxfu93cm5xs/Screenshot%20from%202014-02-18%201 > > 0%3A04%3A03.png > > > > > > I've noticed there are special demosaic algorithms in literature which no > > doubt are sharper and more advanced than my simplistic algorithm. In any > > case I wondered if ppg and amaze are less suited for high-iso noise and we > > might benefit from dedicated noise-capable demosaic code and would > > appreciate a wider range of demosaic code. > > darktable used to offer a wider variation of demosaicing algorithms. We removed > most of them when we rewrote parts of our internal processing pipeline since > no one wanted to write the code and maintain more variants. If you are willing > to provide code for something nifty we can surely negotiate including it. :) > NN interpolation is however not an algorithm I would like to see. It has way > too many issues. > > hanatos, please take over. > > > Best, > > > > Gert > > Tobias > ------------------------------------------------------------------------------ > Managing the Performance of Cloud-Based Applications > Take advantage of what the Cloud has to offer - Avoid Common Pitfalls. > Read the Whitepaper. > http://pubads.g.doubleclick.net/gampad/clk?id=121054471&iu=/4140/ostg.clktrk > _______________________________________________ > Darktable-users mailing list > [email protected] > https://lists.sourceforge.net/lists/listinfo/darktable-users > -------------------------------------------------------------
signature.asc
Description: This is a digitally signed message part.
------------------------------------------------------------------------------ Managing the Performance of Cloud-Based Applications Take advantage of what the Cloud has to offer - Avoid Common Pitfalls. Read the Whitepaper. http://pubads.g.doubleclick.net/gampad/clk?id=121054471&iu=/4140/ostg.clktrk
_______________________________________________ Darktable-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/darktable-users
