Bug #9187

Noise Profile for RX100 not turing out well

Added by Josef Wells over 6 years ago. Updated almost 6 years ago.

Target version:
Start date:
Due date:
% Done:


Estimated time:
Affected Version:
git development version
hardware architecture:


I've tried with two sets of pictures, several times each (with fresh ~/.config/darktable's, no other darktable installed on my system, with fresh git master as of Jan 11 then again Jan 16th).

Noise Profile ARW's:

I ran gen-profile, the curves look mostly ok, the blue noise-level doesn't start near zero (as near zero as red and green), but that is about it.

The .pfm's did not exactly look like the ARW+linearstyle, so I manually generated the pfm's (additionally turning of dithering, which seemed to be on), hacked gen-profile to not regenerate the .pfm's.

The presets do not auto-select the RX100 preset, and when that is selected, it just blackens the picture.

A set of benchmark ARW files is available here, although I am happy to attempt this step again as well:

ISO1600.pdf (26.8 KB) ISO1600.pdf Josef Wells, 02/12/2013 03:00 AM
originalISO1600.pdf (26.9 KB) originalISO1600.pdf Josef Wells, 02/12/2013 04:46 AM
presets.txt (1.92 KB) presets.txt Josef Wells, 02/13/2013 04:41 AM


#1 Updated by Josef Wells over 6 years ago

I tried a new set this weekend and ended up with the same results.

I see now that the linear style is different from the xmp that is applied to generate the pfm files, but either way I get the same results.

The pdf's show an odd curve for blue (fit), that is, nearly linear.. see attached.

#2 Updated by Johannes Hanika over 6 years ago

  • System set to unknown

you shouldn't need to worry about styles and pfms at all.

just copy the raw files in a new, empty directory, and call `gen-profile -d .' from within that.

you might need to do that with a clean ~/.config/darktable/ directory (other user, or move the directory away, run dt once to create the defaults, then run the script).

#3 Updated by Johannes Hanika over 6 years ago

Josef Wells wrote:

The pdf's show an odd curve for blue (fit), that is, nearly linear.. see attached.

the graph looks fine except for the blue channel (which is too dark, your light source will need more power for short wavelengths, or you'll need to overexpose more).

#4 Updated by Josef Wells over 6 years ago

I've tried blowing away ~/.config/darktable and running.

The pdf I added was from the new pictures, not the ones I uploaded earlier, but the pdf's look nearly identical. Both were taken with sunlight in the overexposed areas. (I've added the new 1600ISO image to picassaweb as well)

Attached is the pdf that matches the previously uploaded (picassaweb) pictures.

I'm happy to do more, but I don't know what to try at this point. Please take a peek at the ARW's and pdf's and let me know if you have any suggestions.

#5 Updated by Johannes Hanika over 6 years ago

so do you have a presets.txt for those? i'll just zero out the broken blue channels (unused as of now anyways).

#6 Updated by Josef Wells over 6 years ago

Attached is the presets.txt from the most recent set.

The presets turned my benchmark images (Also on picassaweb) black.

#7 Updated by Johannes Hanika over 6 years ago

great, thanks. i've pushed to master, can you check if that works (sometimes the presets applied in the database have had problems, no idea what's going on there)?

#8 Updated by Josef Wells over 6 years ago

ISO 800 and 3200 don't actually denoise, but the rest work!

I assume the problem is with presets in the database? Anything else I can do?

Here is the run on the (working) pfm results:
iso80_denoised.pfm : 19.8227
iso80.pfm : 19.6832
iso100_denoised.pfm : 19.282
iso100.pfm : 19.0633
iso125_denoised.pfm : 33.1796
iso125.pfm : 30.9525
iso200_denoised.pfm : 25.3061
iso200.pfm : 23.7698
iso400_denoised.pfm : 25.7061
iso400.pfm : 23.014
iso1600_denoised.pfm : 22.7344
iso1600.pfm : 19.1605
iso6400_denoised.pfm : 19.9631
iso6400.pfm : 16.4427

#9 Updated by Josef Wells almost 6 years ago

  • % Done changed from 0 to 100
  • Status changed from New to Fixed

Johannes Fixed this, these noise profiles work well. Thanks!

Also available in: Atom PDF

Go to top