You all are very convincing about highlight reconstruction. There's now an x-trans version of inpaint. It can be a bit noisy in some cases, perhaps due to the less regular sensor pattern, but otherwise looks good.
The main other open question was how/whether to deal with the camera's "Dynamic Range" feature (which turns down the ISO to preserve highlights). For now the global tonemap module's Drago operator seems adequate to make these images look good. Though not clear whether this is a user friendly solution. I madeĀ http://www.darktable.org/redmine/issues/9951 for the x-trans code. Things are far enough along that I could make a pull request, if Darktable developers are amenable? Would it be helpful/useful to clean up the commits, though? (Some are corrections to prior work.) Regardless, I'd be really excited if the Darktable developers had a chance to consider this code, and would be happy to work on fixing problems with it. I'll also keep an eye out for you all on IRC. Thanks, Dan On Sat, May 10, 2014, at 02:04 AM, johannes hanika wrote: > heya, > > nice to see your good progress :) > > one comment on highlight reconstruction. this is necessary because sensor > values are saturated at some same level and then white balance is applied. so > with wb coeffs of (2, 1, 1.8) you would get some magenta highlight instead of > white, in clipped regions. this has nothing to do with bayer patterns at all. > > -jo > > > On Fri, May 9, 2014 at 6:27 AM, Dan Torop <d...@pnym.net> wrote: >> Hello, >> >> The X-Trans sensor branch https://github.com/dtorop/darktable is coming >> along. Lots of work on demosaicing the past few days: >> >> - Markesteijn (high quality) interpolation is better integrated and >> faster. Thanks to pmjdebruijn's suggestion it now uses a Yuv-ish >> perceptual colorspace rather than CIELab, which eliminates a bunch of >> code. Markesteijn comes from dcraw (equivalent to '-q 2' or '-q 3' >> parameters), and does seem to be a modified AHD. >> >> - Added VNG interpolation as the standard x-trans demosaic algorithm >> (it's ok quality and decent speed) -- this is also from dcraw, and the >> same as dcraw's "-q 1" parameter. >> >> Both VNG and Markesteijn now operate on floats (rather than 16-bit ints >> as in the dcraw version) and use OpenMP. >> >> Highlight reconstruction on X-Trans seems unnecessary? Regardless, >> "Reconstruct in Lch" mode does now use OpenMP. The new "reconstruct >> color" mode is disabled for now, though, as it's Bayer dependent. >> >> In off-list exchanges I've heard the branch has worked with files from >> X100S, X-E1, X-T1, and (after a bug fix) X-Pro 1 models. It'd be >> interesting to try it on others. >> >> An open question is whether/how to support Fuji's "Dynamic Range" >> feature, and there are a bunch of ancillary things (UI/config choices, >> optimizations, cacorrect module, etc.). >> >> I'll work on waking up in time to find Darktable developesr on IRC... >> >> Best, >> Dan ------------------------------------------------------------------------------ Is your legacy SCM system holding you back? Join Perforce May 7 to find out: • 3 signs your SCM is hindering your productivity • Requirements for releasing software faster • Expert tips and advice for migrating your SCM now http://p.sf.net/sfu/perforce _______________________________________________ darktable-devel mailing list darktable-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/darktable-devel