On Wed, Apr 27, 2016 at 11:15 AM, johannes hanika <hana...@gmail.com> wrote:
> On Thu, Apr 28, 2016 at 5:13 AM, Ingo Liebhardt <ingo.liebha...@ziggo.nl> > wrote: > > Hi, > > > > In the original message, the 'and have had little trouble generating > images > > that contain areas prone to demosaicing artifacts‘ caught my eye. > > In the meanwhile, I continued a bit on my demosaicing algorithm. > > > > So I gave it a try with the two problematic images. > > Results can be found in: > > > https://www.dropbox.com/sh/un1y11uimbqxjjk/AAD3L-Rs9-ztwyBIm4rnCzK-a?dl=0 > > > > The tiff versions are straight from the algorithm, no correct white > balance, > > no proper contrast curve, no noise reduction. > > The jpeg versions got a bit of white balance and contrast curve in > > darktable, but still no noise reduction. > > Please excuse the colors, my main interest was on the demosaicing, so I > was > > very sloppy on the colors at this stage. > > The tiffs are 16 bit, so you can really play with them to your liking. > > > > Furthermore, I also continued working on a better chroma cleaning > algorithm > > for the fine branches of my own torture test. > > Have a look at the chroma_cleaned_3. I found a way to very specifically > > address those problematic pixels, see map2, which I then further > improved by > > adding a contrast curve to the map. > > > > The bird and person images didn’t get the chroma cleaning, because they > > don’t have fine high contrast areas, unlike the branches. > > > > Jo might be most interested in this, so the concept is in my latest git > > commit : https://github.com/ILiebhardt/xtrans. > > It’s not yet really stable and not to speak of performance… > > > > By the way, I also added my problematic raw, so that Jo can play with it > if > > wanted. > > this is the dt output with just standard processing (xmp see embedded > in jpg; profiled chroma denoising, single pass markesteijn, > classic-chrome-like colours and tonecurve): > > https://jo.dreggn.org/img_0001.jpg > > it doesn't really have the single pixel colour fringes, but i can > still see larger blotches (like 10x10 pixels range maybe?) of green > and magenta. > > cheers, > jo > > You've lost the color of the sunset on the lettering on the clock face though. This is not the best test scene because it doesn't appear to actually contain any vibrant colors. A better scene would have real green, magenta and cyan objects, preferably with fine patterns. That would make it more apparent when the processing is not producing color artifacts vs. just smoothing all of the colors. > > > > I’m working on the performance now. > > > > Cheers, > > Ingo > > > > > > > > Am 27.04.2016 um 18:16 schrieb J. Liles <malnour...@gmail.com>: > > > > > > > > On Wed, Apr 27, 2016 at 6:56 AM, johannes hanika <hana...@gmail.com> > wrote: > >> > >> my 2min-version of the 2nd image: > >> > >> https://jo.dreggn.org/neon.jpg > >> > >> the neon lights are indeed something that doesn't seem to be covered > >> well at all by the reference it8 chart. we probably need to somehow > >> extend the darktable-lut module to read not only well defined colour > >> patches but also random images (as the basecurve tool does). but even > >> then it would just be cloning the jpg engine.. which is sub-optimal > >> imo. i get much improved results when switching on gamut clipping in > >> the input colour profile module (see xmp embedded in the jpg). i > >> suspect the camera manufacturers do similar things on chip. > >> > >> -jo > >> > > > > Looks pretty close. I'd say in the case of Fuji an ability to extract a > > profile from sampling many RAWs+JPGs would probably produce a pretty good > > result. The colors in the camera jpeg were fairly true to life in this > case, > > except a bit desaturated (they were shot using Provia/STD, I've since > > learned that many people believe Fuji mixed up the Provia and Astia > > simulations because Provia seems to have lower saturation than Astia). > But > > as far as the hue goes, it's pretty spot on. Those purple neon lights can > > very easily turn blue. I too have found that enabling gamut clipping > helps > > with that. > > > > > > > > > >> > >> > >> On Tue, Apr 26, 2016 at 10:21 PM, J. Liles <malnour...@gmail.com> > wrote: > >> > Here are some of my problem images. > >> > > >> > http://www.nevermindhim.com/fuji-xtrans-samples > >> > > >> > Also included on the page is my style for denoising. > >> > > >> > > >> > On Tue, Apr 26, 2016 at 11:04 AM, J. Liles <malnour...@gmail.com> > wrote: > >> >> > >> >> Group, > >> >> > >> >> I've recently acquired a camera with an X-Trans sensor, and have had > >> >> little trouble generating images that contain areas prone to > >> >> demosaicing > >> >> artifacts. > >> >> > >> >> Is anyone interested in these? If so, is there a certain place I > should > >> >> host/send them? I have camera generated JPG for comparison along with > >> >> the > >> >> RAF files. > >> >> > >> >> In playing around I've discovered a few things: > >> >> > >> >> 1) Color smoothing is mandatory, as fine lines and especially > specular > >> >> highlights will always generate nasty color artifacts when using the > >> >> baisc > >> >> demosaicing. Seems like a smarter highlight recovery algorithm could > >> >> help > >> >> identify many of these problem spots. > >> >> > >> >> 2) None of the 'denoise' modules have sufficient parameter ranges to > >> >> deal > >> >> well with the X-Trans files. > >> >> > >> >> 3) When sufficient denoising is performed to remove maze and color > >> >> artifacts, the difference between VNG and Markesteijn demosaicing is > >> >> indiscernible. > >> >> > >> >> I have a style preset which can produce images that match the camera > >> >> JPGs > >> >> as far as noise/detail goes (utilizing the equalizer module) at ISO > >> >> 3200. > >> >> > >> >> Color is another matter, but I've ordered one of the Wolf Faust IT8 > >> >> charts > >> >> to try and make an ICC profile. > >> >> > >> >> (I tried making one using the imaging-resource.com multi target > studio > >> >> shots, and while the results were OK, they weren't great). > >> >> > >> >> P.S. Is there any plan to support Fuji's RawExposureBias exif tag? It > >> >> is > >> >> necessary to display high ISO RAF images at the correct brightnesss. > >> > > >> > > >> > > >> > > >> > > ___________________________________________________________________________ > >> > darktable developer mailing list to unsubscribe send a mail to > >> > darktable-dev+unsubscr...@lists.darktable.org > > > > > > > > > ___________________________________________________________________________ > > darktable developer mailing list to unsubscribe send a mail to > > darktable-dev+unsubscr...@lists.darktable.org > > > > > ___________________________________________________________________________ darktable developer mailing list to unsubscribe send a mail to darktable-dev+unsubscr...@lists.darktable.org
