On 04/13/15 06:09, Tobias Ellinghaus wrote:
> Am Montag, 13. April 2015, 02:02:22 schrieb Urs Schütz:
>> I did benchmark a camera noise profile [1]. My input files [2] go from
>> ISO 100 to ISO 25600, at powers of 2 (100,200,400,...,12800,25600).
>>
>> To prepare the benchmarking I did the following within darktable
>> 1.7.0+1256~g960e920:
>> Create a hdr out of five images at camera base ISO 200.
>> Create two minimal styles: WithNR (profiled denoise ON) [3] and NoNR [4]
>> (profiled denoise OFF).
>> Discard the history stack of all test images.
>> Select all images and double-click NoNR profile.
>> Export all images as PFM(float), using the NoNR profile, replace history.
>> Discard the history stack of all test images.
>> Select all images and double-click WithNR profile.
>> Export all images as PFM(float), using the WithNR profile, replace history.
>>
>> Here comes the problem:
>> The benchmark results [5] show a strong increase of signal to noise
>> ratio for part of the images, but the three images at ISO 1600, 3200 &
>> 6400 show only minimal or no increase of the signal to noise ratio.
>>
>> On visual inspection, there was no big change in noise appearance for
>> this three images. Others worked better or excellent (as usual with the
>> very good profiled denoise module).
>>
>> In darkroom mode I checked the profiled denoise image by image, and
>> found that the profiled denoise module is using wrong profiles for some
>> of the images.
>> ISO 400, 800, 1600, 3200, 6400 has wrong automatic selection of denoise
>> profile ("using interpolated from ISO 100 and ISO 12800").
>> For ISO 100, 200, 12800, 25600, the selected denoise profile is as
>> expected ("found match for ISO ...").
>>
>> After selecting the matching denoise profiles manually from the
>> drop-down list, I get much better signal to noise ratio [5], and
>> visually much cleaner files.
>>
>> The used profile presets.json [1] contains entries for all nine ISO
>> settings.
>>
>> Why does the profiled denoise module partially select interpolated
>> profiles, even when there are matching profiles? Does this just happen
>> with my denoiseprofile [1], or is it happening with other
>> denoiseprofiles also?
>
> Thanks for reporting, this should be fixed in git master now. The problem was
> that the noise profiles in your json file are not ordered by ISO.
>
>> Urs
>
> Tobias
>
>> [1]
>> https://www.dropbox.com/s/e7we55r8ru5p14h/dt-noiseprofile-20150412.tar.gz?dl
>> =0 [2] https://www.dropbox.com/s/w8wqjm8bzsltawi/RawTestImages.tgz?dl=0 [3]
>> https://www.dropbox.com/s/tg6izewjrcqsuex/NoNR.dtstyle?dl=0
>> [4] https://www.dropbox.com/s/1v6txb7b1sk82v7/withNR.dtstyle?dl=0
>> [5] https://www.dropbox.com/s/9y37i3ev507b1mo/Benchmark%20Results.ods?dl=0Automatic profile detection is working correctly now. Thanks a lot for your fast help! Urs ------------------------------------------------------------------------------ BPM Camp - Free Virtual Workshop May 6th at 10am PDT/1PM EDT Develop your own process in accordance with the BPMN 2 standard Learn Process modeling best practices with Bonita BPM through live exercises http://www.bonitasoft.com/be-part-of-it/events/bpm-camp-virtual- event?utm_ source=Sourceforge_BPM_Camp_5_6_15&utm_medium=email&utm_campaign=VA_SF _______________________________________________ Darktable-users mailing list Darktable-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/darktable-users
