On Thu, 8 Dec 2016 12:08:52 +0100, Stéphane Gourichon <[email protected]> wrote :
> Le 08/12/2016 à 00:44, Saint Germain a écrit : > > Hello, > > > > I just bought a Samsung NX1 and would like to use Darktable to > > process the RAW (SRW). > > > > However I really tried but I couldn't manage to get correct colors > > or to have a good noise reduction in high iso situation (low light). > > > > I've tried LightRoom and RawTherapee and I managed (without knowing > > these softwares) to get quite good results in under 5-10 minutes. > > So far I think I've spent several hours with Darktable... > > > > I would really like to stay with Darktable (as I find the interface > > quite pleasant and it offers a lot of useful features) but I really > > need to have good colors and good noise reduction (at least > > comparable to LR and RT) without too much tinkering. > > > > From what I have heard, it seems that Darktable noise reduction is > > very powerful and actually superior to LR or RT. Basic noise > > reduction in LR or RT is very simple : 2 sliders and that's it > > (Luminance and Details). However it is quite effective, and I > > haven't been able to do better with Darktable (and it took me much > > more time, I even made a noise profile !). > > > > Perhaps am I doing something wrong ? > > Can someone guide me or show me how to proceed ? > > Hello. > > Thanks for sharing all details. > Attached is my take on it. > > Given base material (high iso, not much actual detail) I took a > somehow more aggressive approach than usual (on the "preserve detail" > vs. "kill noise" spectrum). > > Here's the parameter dump: > * disable base curve > * disable profiled denoise > * RAW noise reduction, default parameters > * increase exposure +3.10EV so that brightest part (yellow title) is > just on top > * bilateral filtering, red 0.04 green 0.02 blue 0.1 > * non-local means, default parameters > * optionally, equalizer preset "noise reduction" > * sharpen, default parameters > > Here are the rationales for those choices : > * disable base curve: source is low dynamic range (paper) and > furthermore already has its own base curve > * disable profiled denoise: haven't imported your profile. Anyway, > when strong, generally does not perform well with bilateral. > * RAW noise reduction: filters part of noise, at the cost some > high-frequency killing which is okay in this specific case > * bilateral filtering: attenuates large area hue variations somehow, > without losing details at this strength > * non-local means: filters out some residual speckles > * equalizer: reduces noise some more, at the cost of more loss of > details > * sharpen: to selectively recover edges attenuated by equalizer > > Somehow off-topic: > * to correct barrel: chosen hackish lens correction parameters: NX10 > and 20-50 F3.5-5.6 ED, looks good > * set white balance "spot" selecting white part of title. Much better > color, well, balance. Reset it to see how previous case could hide > yellowish noise stains in white areas. > * Subject is still. Why didn't you choose a longer exposure ? ISO > would have been lower, so less noise to begin with. > Thanks for this attempt and the detailed answer ! Please not that I uploaded the wrong noise profile by mistake, here is the correct one: http://dl.free.fr/bNDlCNHWH The result you got is Ok but in my opinion a little less efficient from what I managed to obtain with LightRoom (or Rawtherapee). If you look at the woman on the left, espcially the hair above her ear and the cheek, it seems that we have lost a lot of details. I'll try to play a little with the parameters you gave to see if I manage to get something better. * lens correction: I have to go out and find a nice building to make the lens correction for lensfun * white balance can indeed be corrected. However why is LightRoom and Rawtherapee giving me directly the correct WB ("Body") whereas Darktable need some addition post-processing ? * The use case was to shoot children inside my appartment. To simulate that, I took a picture of a magazine ! ;-) However I tried with the real children and the results are the same. ____________________________________________________________________________ darktable user mailing list to unsubscribe send a mail to [email protected]
