thanks, not sure what kind of help i would ask for at this point. i noticed that your flattened histogram even with a slightly improved fitting procedure that i just tested (will push later, doesn't really help all that much so far) has a 2x noise std deviation for dark regions. some other cameras have way worse behaviour than that for brightnesses < 0.2, but maybe this is enough to confuse the algorithm. i fitted a gaussian to that error and instructed the wavelet denoiser accordingly, but so far results aren't a lot better than what we had before.
i could get way better results using the equalizer and manually pulling up the threshold for the chroma channels (start from `chroma denoise' preset and push it to the extreme basically). i hope to do some more tests on that tomorrow. j. On Tue, May 28, 2013 at 6:21 AM, Baptiste Pierrat < [email protected]> wrote: > I´m willing to help if you want more samples or else (I´m not a good coder > but I have some basic knowledge). > > > On Mon, May 27, 2013 at 2:50 PM, johannes hanika <[email protected]>wrote: > >> hm. if anything your sensor is cooler now and has less constant noise >> added on top. the data in your image matches your previous measurements >> very closely. i'll have to do some more digging. >> >> j. >> >> >> On Mon, May 27, 2013 at 1:46 PM, Baptiste Pierrat < >> [email protected]> wrote: >> >>> Well I did the profiling for my camera and it's currently this one that >>> is used in the code (git master). I took great care in doing this >>> correctly, I did 4 profiles and averaged the results. >>> >>> Here is the raw file from the previous example if you want to test: >>> https://www.dropbox.com/s/iqb04gly913vmfx/20130429_0041.ARW >>> >>> The problem for me is not only the green pixels but also the bigger >>> cross-shaped blue and red dots. Isn't that somehow related to the >>> demosaicing process? >>> >>> Thank you for helping with that. >>> >>> >>> >>> >>> On Mon, May 27, 2013 at 1:27 PM, johannes hanika <[email protected]>wrote: >>> >>>> hm. that looks reasonable. very surprising that those green pixels >>>> still stay in the color buffer. maybe the profiling doesn't match your >>>> particular camera model? maybe the profile was a little off for that >>>> particular iso value, or maybe there is some variance between individual >>>> sensors for sony cams? >>>> >>>> i can try and re-profile that particular raw and see if there are any >>>> differences to the data we have if you want. >>>> >>>> j. >>>> >>>> >>>> On Mon, May 27, 2013 at 12:14 PM, Baptiste Pierrat < >>>> [email protected]> wrote: >>>> >>>>> [image: Inline image 1] >>>>> >>>>> >>>>> On Mon, May 27, 2013 at 12:12 PM, johannes hanika >>>>> <[email protected]>wrote: >>>>> >>>>>> what are the white balance coefficients for that image? >>>>>> >>>>>> >>>>>> On Mon, May 27, 2013 at 11:57 AM, Baptiste Pierrat < >>>>>> [email protected]> wrote: >>>>>> >>>>>>> Hello, >>>>>>> >>>>>>> I want to second this problem. >>>>>>> >>>>>>> I have a Sony Alpha 57 and have a lot of trouble with remaining >>>>>>> bright color dots after denoising. My usual denoising tools for iso>800 >>>>>>> are >>>>>>> : >>>>>>> 1. Chroma denoise: profiled denoising with wavelets, blend with >>>>>>> color, then >>>>>>> 2. Luma denoise: profiled denoising with non-local means, blend with >>>>>>> lightness >>>>>>> >>>>>>> Here is an example with an iso 3200 sample: >>>>>>> original: http://minus.com/lU783RXDLcLH7 >>>>>>> chroma denoised: http://minus.com/lBe1kgj6o5aoP >>>>>>> chroma and luma denoised: http://minus.com/lbqRsDO5q4E617 >>>>>>> detail after export: http://minus.com/lCLmKR1s2vh5a >>>>>>> >>>>>>> Green, red and blue larger dots remain. Also notice as chroma >>>>>>> denoising does not work well for green color. As suni, this problem is >>>>>>> partially solved by using the color smoothing option in demosaic but the >>>>>>> image is slightly blurred. >>>>>>> >>>>>>> Thanks! >>>>>>> >>>>>>> >>>>>>> On Mon, May 27, 2013 at 11:25 AM, johannes hanika <[email protected] >>>>>>> > wrote: >>>>>>> >>>>>>>> you left in all the luma noise in your lightroom comparison image. >>>>>>>> to do that in darktable, use `color' blend mode in the profiled >>>>>>>> denoising >>>>>>>> module. you may want to use the wavelet method, it is usually a little >>>>>>>> more >>>>>>>> effective against color noise. >>>>>>>> >>>>>>>> j. >>>>>>>> >>>>>>>> >>>>>>>> On Mon, May 27, 2013 at 11:21 AM, suni <[email protected]> wrote: >>>>>>>> >>>>>>>>> Hi all, >>>>>>>>> >>>>>>>>> I'm fighting the high ISO noise of my A550 camera, but I still >>>>>>>>> cannot achieve the reasonable output. Even the new profiled denoise >>>>>>>>> filter >>>>>>>>> doesn't help much. >>>>>>>>> >>>>>>>>> Below is the example of the ISO12800 image. I do not take images >>>>>>>>> on such high ISO but it makes this problem more visible for this >>>>>>>>> purpose. >>>>>>>>> This behaviour may be observed even at ISO1600 and higher. The RAW >>>>>>>>> image >>>>>>>>> below is totally unaltered RAW for comparison opened side by side in >>>>>>>>> dt and >>>>>>>>> in Lt. Only the base curve is applied at dt side because otherwise the >>>>>>>>> image looks very ugly and not comparable to the Lt from the visual >>>>>>>>> point of >>>>>>>>> view. >>>>>>>>> >>>>>>>>> Link to unaltered RAW image: >>>>>>>>> https://dl.dropboxusercontent.com/u/9454436/sony_noise/raw_not_altered.jpg >>>>>>>>> >>>>>>>>> As you can see the image looks almost the same in both tools. Just >>>>>>>>> note the bright red dots. The Lt shows less of them especially on the >>>>>>>>> right >>>>>>>>> side of the paper Christmas tree. These red dots cannot be removed by >>>>>>>>> any >>>>>>>>> of the dt denoising filters unless the image becomes almost blurred. >>>>>>>>> >>>>>>>>> Now see the next image below. This shows what can be achieved in >>>>>>>>> dt and in Lt. While only activating the luma and chroma denoise >>>>>>>>> sliders in >>>>>>>>> Lt produces the acceptable and (to me) pretty usable output where >>>>>>>>> only some >>>>>>>>> luma noise remains, with dt I cannot achieve similar output by no >>>>>>>>> means. >>>>>>>>> The most strange finding is that no one from all the denoising filters >>>>>>>>> available in dt can remove these bright red dots unless it is pushed >>>>>>>>> to not >>>>>>>>> acceptable high strength producing the blurred image. The only >>>>>>>>> strategy I >>>>>>>>> found to be somehow working here is as follows: >>>>>>>>> 1) turn off the sharpen filter >>>>>>>>> 2) apply the hot pixel filter at its maximum strength >>>>>>>>> 3) set the color smoothing of the demosaic filter to maximum >>>>>>>>> >>>>>>>>> These three steps do remove the bright red dots. But the image >>>>>>>>> still suffers of the high chroma noise. The only possible way to get >>>>>>>>> rid of >>>>>>>>> it while maintaining the image sharpness at the level comparable to >>>>>>>>> the Lt >>>>>>>>> I found is the profiled denoise set to less than default strength. >>>>>>>>> But as >>>>>>>>> you can see there still remain far more chroma noise than in the Lt >>>>>>>>> case. >>>>>>>>> >>>>>>>>> Link to altered RAW image: >>>>>>>>> https://dl.dropboxusercontent.com/u/9454436/sony_noise/raw_altered.jpg >>>>>>>>> >>>>>>>>> Because I've got to that point I didn't know what to do more, >>>>>>>>> I profiled my camera. But even though my profile seems to me to be a >>>>>>>>> bit >>>>>>>>> better than the one packed with dt, it produces no visible difference >>>>>>>>> here. >>>>>>>>> >>>>>>>>> I would like to ask all of you experienced in fighting the high >>>>>>>>> ISO noise for your experiences and recommendations. It would be great >>>>>>>>> to >>>>>>>>> get to know that the problem is in between the chair and the >>>>>>>>> keyboard, not >>>>>>>>> in the excellent dt tool!:-) >>>>>>>>> >>>>>>>>> If someone is interested, the original RAW image can be found >>>>>>>>> here: >>>>>>>>> https://dl.dropboxusercontent.com/u/9454436/sony_noise/DSC09533.ARW >>>>>>>>> >>>>>>>>> >>>>>>>>> Thank you in advance for any reply! >>>>>>>>> >>>>>>>>> Suni >>>>>>>>> >>>>>>>>> >>>>>>>>> ------------------------------------------------------------------------------ >>>>>>>>> Try New Relic Now & We'll Send You this Cool Shirt >>>>>>>>> New Relic is the only SaaS-based application performance >>>>>>>>> monitoring service >>>>>>>>> that delivers powerful full stack analytics. Optimize and monitor >>>>>>>>> your >>>>>>>>> browser, app, & servers with just a few lines of code. Try New >>>>>>>>> Relic >>>>>>>>> and get this awesome Nerd Life shirt! >>>>>>>>> http://p.sf.net/sfu/newrelic_d2d_may >>>>>>>>> _______________________________________________ >>>>>>>>> Darktable-users mailing list >>>>>>>>> [email protected] >>>>>>>>> https://lists.sourceforge.net/lists/listinfo/darktable-users >>>>>>>>> >>>>>>>>> >>>>>>>> >>>>>>>> >>>>>>>> ------------------------------------------------------------------------------ >>>>>>>> Try New Relic Now & We'll Send You this Cool Shirt >>>>>>>> New Relic is the only SaaS-based application performance monitoring >>>>>>>> service >>>>>>>> that delivers powerful full stack analytics. Optimize and monitor >>>>>>>> your >>>>>>>> browser, app, & servers with just a few lines of code. Try New Relic >>>>>>>> and get this awesome Nerd Life shirt! >>>>>>>> http://p.sf.net/sfu/newrelic_d2d_may >>>>>>>> _______________________________________________ >>>>>>>> Darktable-users mailing list >>>>>>>> [email protected] >>>>>>>> https://lists.sourceforge.net/lists/listinfo/darktable-users >>>>>>>> >>>>>>>> >>>>>>> >>>>>> >>>>> >>>> >>> >> >
<<image.jpeg>>
------------------------------------------------------------------------------ Try New Relic Now & We'll Send You this Cool Shirt New Relic is the only SaaS-based application performance monitoring service that delivers powerful full stack analytics. Optimize and monitor your browser, app, & servers with just a few lines of code. Try New Relic and get this awesome Nerd Life shirt! http://p.sf.net/sfu/newrelic_d2d_may
_______________________________________________ Darktable-users mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/darktable-users
