Hi Florian, sooo, I checked your files and can confirm your findings.
The first issue (off pixels) is obviously a bug. As you already wrote we are converting values from RGB to HSL and back. If the input is out of range, this leads to problems. In current git master I fixed the issue by clamping the input values accordingly (which btw was already done in our OpenCL path). I might need to inspect further to find out where the out-of-range pixels are generated. It's quite clear we *need* to clamp all input pixels in blending if we want to transfer to some other color space. The second issue with equalizer might be an inherent property of that module. Older versions of darktable show the same effect. In fact similar settings lead to comparable artifacts also on other images that I have tested. So you probably went over the top with your choice of module values :) Ulrich Am 17.01.2013 07:45, schrieb Florian GOULEAU: > Hi Ulrich, > > i made a zip with both RAW and XMP : > > http://dl.free.fr/b9aJ2E9m2 > > > Florian > > > Le 17/01/2013 07:32, Ulrich Pegelow a écrit : >> Hi, >> >> both cases would deserve further inspection. The first issue might be >> related to hot pixels but it might also be related to our way of >> calculating from RGB to HSL and back - which is done to process blend >> mode lightness. >> >> The second issue is even a bit more strange. I often use equalizer for >> local contrast enhancement myself, but never saw these repeated pattern >> of image features. >> >> We will need your RAW files and XMP files. Maybe you can upload them >> somewhere. >> >> Ulrich >> >> Am 17.01.2013 00:34, schrieb Florian GOULEAU: >>> Dear all, >>> >>> Let me first whish you all a happy and fruitful year 2013. >>> 2012 was the year of maturity for Darktable with the release of 1.0 >>> and the subsequent updates. May 2013 be a year with a lot of great >>> and expected new capacities (masks ?). >>> >>> The addition of the new denoise module in december was an early start. >>> It gives amazing results with my old EOS 350D and new perspectives in >>> the usage of high ISO (ie 1600 ISO for the 350D !). >>> >>> >>> However, while testing this new module, I came upon two glitches, not >>> related to the module, for which I would like your opinion : >>> >>> 1. >>> When applying the denoise non local means blended at 75% in lightness >>> mode (as proposed by Johannes), I do have many colored pixels in the >>> darkest areas of my photo. Is that an expected behavior when using >>> lightness blending ? >>> (see the following crop and the full picture linked to this image) >>> >>> <http://farm9.staticflickr.com/8471/8384177345_df88c691fb_o.png> >>> >>> >>> >>> 2. >>> When using the global contrast preset of the equalizer module, I observed >>> a pattern repeated on the image. And curiously, when applying a smaller >>> crop to export the example below, it produced a pattern which is different >>> from the one on a larger crop or the full image. >>> Are both the patterns and the crop dependency expected when using >>> equalizer ? >>> (see the following crop and especially the larger picture linked to this >>> image) >>> >>> <http://farm9.staticflickr.com/8221/8384178767_235fbe9ace_o.jpg> >>> >>> >>> >>> >>> Last question, I have seen in the noise profiles that you stored >>> complementary >>> profiles in order to eventually average the different profiles one day. >>> Do you expect us to provide additional noise profile results even if our >>> camera >>> is already covered ? >>> >>> >>> cheers, >>> >>> Florian ------------------------------------------------------------------------------ Master HTML5, CSS3, ASP.NET, MVC, AJAX, Knockout.js, Web API and much more. Get web development skills now with LearnDevNow - 350+ hours of step-by-step video tutorials by Microsoft MVPs and experts. SALE $99.99 this month only -- learn more at: http://p.sf.net/sfu/learnmore_122812 _______________________________________________ darktable-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/darktable-devel
