[argyllcms] Re: Argyll 1.1.0 RC2 luminance readings high by ~4-6% with Eye-One Pro

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Mon, 21 Dec 2009 12:50:47 +1100

Alexander wrote:
I'm having an issue where the luminance readings from Argyll 1.1.0 RC2
are always high by about 4-6% when compared to any other application
that uses the EyeOne.dll SDK provided by X-Rite.

Hmm. I checked using my Rev D., and got a discrepancy of about 1%
at a luminance level of about 42 cd/m^2, which (last time I looked
into this) seemed close enough, given that I wasn't sure that the
reference (my display) was more stable than that. Looking at it again
however, I have found that this discrepancy is explained by a bug
in my driver - I've accidentally swapped the high and low gain
linearity factors :-( :-( :-(
Fixing this up and testing again over an average of 5
readings of white with both drivers gives an error of 0.05%,
an improvement of 20! For my Rev A instrument the agreement
between the two drivers has a discrepancy of 0.02%.

So it will be interesting to see if your error goes away
with this fix.

[I note though that there is a 12% discrepancy between
 the absolute values of the Rev A and Rev D instruments I have
 access to - absolute calibration is notoriously difficult.]

Graeme Gill, on a somewhat unrelated note could you add an option to
dispcal and dispread to always average by a user specified (or
automatic?) number of readings before recording/using/displaying any
such value? Adding a threshold value to the previous idea may not be a
bad idea either.

I'm not so sure about this. dispcal is already very, very slow compared
to other calibration software. Internally the i1pro driver
already checks a reading for consistency over the integration period.
I understand that you are seeing some "interesting" behavior in
your particular situation, but before adding some sort of counter measure,
it would be better to understand what is causing such inconsistency.

Since the values read jump around slightly on my CRT, especially for
dark readings (probably because of the Eye-One Pro which I've heard is
weak at low luminance),

The combination of a CRT display and the i1pro is not as refined as some
of the dedicated colorimeter such as the DTP92 or DTP94 which have
hardware synchronization to the CRT refresh period. All you can
do with the i1pro is to use a long enough integration time to make
the refresh influence small. A longer integration time also improves
low light accuracy. A more interesting option might be
to have a mode for dispcal and dispread to use the i1pro adaptive
reading mode rather than a fixed integration time. This should
result in longer integration times for darker values.

Graeme Gill.

Other related posts: