Project

General

Profile

Camera Support #11413

noise profile for Panasonic DMC G8*

Added by Andreas Schleth over 2 years ago. Updated about 2 years ago.

Status:
Fixed
Priority:
Low
Assignee:
-
Category:
-
Target version:
Start date:
01/02/2017
Due date:
% Done:

100%

Affected Version:
2.2.0

Description

I followed the steps in http://www.darktable.org/2012/12/profiling-sensor-and-photon-noise/ and produced a noise profile.
The profile was prepared with a G81 and is already corrected to work with G8*.

The benchmark results seem to be OK:
ISO original denoised improvement
200 48,3991 49,8266 2,95 %
400 41,395 42,2368 2,03 %
800 38,4548 39,3159 2,24 %
1600 38,8393 40,956 5,45 %
3200 36,9439 40,0253 8,34 %
6400 35,8635 42,6279 18,86 %
12800 32,687 39,5521 21,00 %
25600 29,098 35,5539 22,19 %

At low ISO values (200 is the lowest for this camera), the profile tends to eat up small features (eg. leaves and branches in a panorama of woodland).
At (very) high ISO values, the profile seems to do not much of an improvement (on visual inspection). If I push up the strength, I end up with a blotchy image. The wavelet method seems to improve on this but at the same time tends to eat up a lot more detail.

The benchmark RW2-images are in my dropbox folder: https://www.dropbox.com/sh/tcgz5oc1bhm6yxp/AAAzDrVynQ0DVXg5jYTbpLENa?dl=0
Please complain, if downloading fails.

dt-noiseprofile-20170101.tar.gz - noise profile info Panasonic DMC G8 (G81) (10.2 MB) Andreas Schleth, 01/02/2017 11:09 AM

Associated revisions

Revision 14522bca
Added by Stefan Schöfegger over 2 years ago

noise profile: add Panasonic DMC G8, fixes #11413, fixes #11305

used data from #11413, looks better than #11305

Revision d70cfa4f
Added by Roman Lebedev over 2 years ago

Merge pull request #1401 from schenlap/noise_panasonic_dmc_g8

noise profile: add Panasonic DMC G8, fixes #11413, fixes #11305

Revision dbda27ac
Added by Stefan Schöfegger over 2 years ago

noise profile: add Panasonic DMC G8, fixes #11413, fixes #11305

used data from #11413, looks better than #11305

(cherry picked from commit 14522bca5005c5909cb9823a3401ecf9b308c474)

History

#1 Updated by Andreas Schleth over 2 years ago

Sorry, the benchmark table was wrong (raw linear profile on output). Here is what happens without an output profile:

ISO original denoised improvement
200 25,1359 25,2732 0,55 %
400 24,0828 24,4514 1,53 %
800 23,036 23,6608 2,71 %
1600 22,8612 24,1667 5,71 %
3200 21,6237 23,9141 10,59 %
6400 20,4031 24,7765 21,43 %
12800 17,9513 24,0032 33,71 %
25600 15,305 22,5363 47,25 %

#2 Updated by Stefan Schöfegger over 2 years ago

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

#3 Updated by Raphaël Monat over 2 years ago

Hi,

Thank you for the profile. Is there any reason to use a 1 EV ISO increment (rather than 1/3) in the noise profile?

#4 Updated by Andreas Schleth over 2 years ago

Hi Raphaël,

I only saw your profiles after I added my own. Sorry to duplicate your work. I hope the profiles are somewhat similar - I have to look at the files.

First, I am quite new to this camera, so I did not yet find the "extended ISO" settings (the menus are quite daunting) for 100 ISO.
Second, it takes quite a bit of fiddling to get acceptable shots for the noise profile generator tool.
Third, I do not think too much accuracy is really necessary here, as other factors are quite important as well:
  • if you diddle with exposure in DT, +1 EV in exposure is the equivalent of changing the ISO by +1 EV in terms of noise (see the test at dpreview). The automatic profile does not take this into account.
  • the profile automatic with standard settings is quite OK if you take images of large plastic parts. For faces (hair) and other hairy things (think woodland panoramas etc.) the standard settings take out too much detail, making the images look like shots from a 80€-camera.
  • for really large ISO numbers, there remains a blotchy pattern in standard settings. This is a bit better with the wavelet method, but there you need to turn down the strength eventually or you kill all detail.
  • did I not read, "we can do interpolation" or something along these lines?

For now I use my profile with strength 0.2 and wavelet for iso200-400 and add a little bit of grain (~10%) afterwards to get rid of the "plastic/painted" look. In effect this takes away the colour noise and adds brightness noise. for higher iso-numbers I increase the factor gradually 0.3 / 0.35 / 0.4 for 800 / 1600 / 3200. However, I try to avoid larger isos as much as possible.

NB: is there a better place to do discussions like this?

Andreas

#5 Updated by Roman Lebedev over 2 years ago

  • Target version set to 2.4.0

#6 Updated by Raphaël Monat about 2 years ago

Hi Andreas,

Sorry for the late answer, I have not been notified. I think your profiles are better, thank you for your input. I will definitely try the values you mention, and I agree on the fact that avoiding larger isos is still the best thing to do for now.

Raphaël

Also available in: Atom PDF