Project

General

Profile

Camera Support #10433

Better white levels for canon 100D

Added by Adrien Descamps almost 5 years ago. Updated over 3 years ago.

Status:
Fixed
Priority:
Low
Assignee:
-
Category:
Darkroom
Target version:
Start date:
04/26/2015
Due date:
% Done:

100%

Estimated time:
Affected Version:
git stable branch

Description

Hi,

Please find attached a patch with more accurate white levels for Canon 100D.

Note that the procedure described at http://www.darktable.org/2012/10/whats-involved-with-adding-support-for-new-cameras/ did not give accurate results.

I took a few fully overexposed pictures with different iso values and looked at the mean raw value (computed using rawspeed).
There are slight variations, but globally levels are very close to the levels defined for the canon 700D, so i used the same levels for consistency. The only difference with the 700D is the iso_min of the high iso range, which seems to be 12800 instead of 6400.
I guess that the levels of the 700D were determined using a similar method.
Maybe it would be worth to document it for other users? If you are interest, let me know how i may help.

Best,

Adrien

0001-better-white-levels-for-canon100D.patch (951 Bytes) 0001-better-white-levels-for-canon100D.patch Adrien Descamps, 04/26/2015 08:13 PM
white_levels.txt (374 Bytes) white_levels.txt Adrien Descamps, 04/27/2015 10:45 PM

Associated revisions

Revision 0896183a (diff)
Added by Roman Lebedev over 3 years ago

Canon 100D: proper whitelevels. Fixes #10433

Revision 415f3c9f (diff)
Added by Roman Lebedev over 3 years ago

Canon 100D: proper whitelevels. Fixes #10433

(cherry picked from commit 0896183ad38765354215645bb80cb40f4cc67037)

History

#1 Updated by Pascal de Bruijn almost 5 years ago

What do you mean by the procedure gave inaccurate results? Presumably you just need to do that with DNG converted samples for all ISO levels.

This script might help a bit:
https://raw.githubusercontent.com/darktable-org/darktable/master/tools/dngmeta.sh

#2 Updated by Adrien Descamps almost 5 years ago

I mean that values were significantly different from what i found using overexposed pictures (DNG Converter gives 12277 for 100 iso and 15000 for other iso values). I used last version of DNG Converter, which officialy support Canon 100D

I may be doing something wrong, but using my values, histograms of overexposed picture (with all modules disabled) look correct for all iso levels.
Also, the fact that my results are consistent with some others canon camera models in cameras.xml makes me think it is correct, or at least that someone else is doing the same thing than me.

By the way, sorry for the wrong tracker, this obviously must go to "Camera support", not "Bug", but it seems that i can't change it myself.

#3 Updated by Pascal de Bruijn almost 5 years ago

Can you supply us with a list of white levels per ISO (from the DNGs), so we can see what values you're seeing?

#4 Updated by Adrien Descamps almost 5 years ago

You will find attached a table with the white levels i have.

To be more clear, here is what i did:
I took two series of white pictures with very long exposure time (30'') at each iso value.
The 'dng metadata' row are the value given by the DNG Converter method on the first series.
The 'white1' and 'white2' are the value given by computing the mean of raw values of each series, with a small program using rawspeed.
There is one missing picture (400iso) in second series that was accidentally deleted.

The mean values were computed from CR2 files, not DNG, and i just noticed that it makes a difference since rawspeed does not take the cropping into account for CR2 files.
However, the difference in mean value is really marginal (less than one unit).

You will find all data (including program source, and another series of pictures with overexposed area) here:
https://app.box.com/s/zop2ai8jawdfr4gf65gen1phj3l4l240

#5 Updated by Pedro Côrte-Real almost 5 years ago

Could you please attach the raw files you got these from so we can have a look?

#6 Updated by Pedro Côrte-Real almost 5 years ago

Sorry, just realized you had added a link to the files already.

#7 Updated by Pedro Côrte-Real almost 5 years ago

Setting the values reported by the DNG looks fine to me. Here are the lines I used:

  <Sensor black="0" white="12277" iso_min="0" iso_max = "199"/>
  <Sensor black="0" white="15000"/>

#8 Updated by Adrien Descamps almost 5 years ago

Ok, it just didn't seem right to me because those values truncate more than 10% of the range for some iso.
However, if you think it is not a problem in practice and your policy is to use the values from DNG converter, go ahead.

#9 Updated by Pedro Côrte-Real almost 5 years ago

We'd like to set whatever is best. I checked your program and indeed we seem to be getting >15000 values. Don't know why you're using the mean and not the max values though. Looking at this it seems we're probably better off leaving it as it is and letting the white point go to 16383 to avoid any clipping.

#10 Updated by Adrien Descamps almost 5 years ago

The problem with the white point at 16383 (and generally, a too high value), is that it results in a pink color in overexposed area.
The effect is very subtle at 200-6400 iso, but clearly visible at 100 iso. Actually, this is probably worst than clipping a part of the histogram for most users, since the problem is more visible.

I used the mean value because the max values seems slightly less stable and are probably sensitive to sensor noise. Moreover, means values are similar to those of other models found in cameras.xml (eg 700D), which made me more confident that my method was correct.

#11 Updated by Pedro Côrte-Real over 4 years ago

  • Tracker changed from Bug to Camera Support

We lost track of this discussion. Did you come to any conclusion on what the best settings are?

#12 Updated by Pedro Côrte-Real about 4 years ago

  • % Done changed from 70 to 20
  • Status changed from Patch attached to Incomplete

#13 Updated by Roman Lebedev over 3 years ago

  • % Done changed from 20 to 100
  • Status changed from Incomplete to Fixed

#14 Updated by Roman Lebedev over 3 years ago

  • Target version set to 2.2.0

Also available in: Atom PDF

Go to top