[argyllcms] Re: Dispcal - Inverting Jacobian failed?

Alastair M. Robinson wrote:
> I did take a few readings (with a DTP94, linear LUT and the same
> monitor settings as had led to this error message) with a target
> comprising just the top end of the greyscale - here are the readings:
>
> NUMBER_OF_SETS 12
> BEGIN_DATA
> 1 95.276 95.276 95.276 89.871 98.192 136.72
> 2 95.669 95.669 95.669 90.191 98.443 136.52
> 3 96.063 96.063 96.063 90.544 98.826 136.01
> 4 96.457 96.457 96.457 90.883 99.170 135.39
> 5 96.850 96.850 96.850 91.320 99.484 135.21
> 6 97.244 97.244 97.244 91.659 99.676 135.06
> 7 97.638 97.638 97.638 92.180 99.980 135.06
> 8 98.031 98.031 98.031 92.209 99.985 135.05
> 9 98.425 98.425 98.425 92.200 99.971 135.04
> 10 98.819 98.819 98.819 92.209 99.980 135.05
> 11 99.213 99.213 99.213 92.200 99.990 135.03
> 12 99.606 99.606 99.606 92.214 99.985 135.06
> END_DATA
>
> Note some reversals in there.

There is obviously a "clipping knee" at about 97% RGB (with a flat
region beyond the knee), but even more interesting is the negative
dX/dRGB gradient in the non-clipped region. I'm indeed wondering how
this can happen - maybe by applying a matrix (with nagative entries) to
non-linear or already clipped RGB numbers.

Btw, also conspicuous is the excellent repeatability of this instrument
(in the flat region, readings #7...#12) - I don't get such low noise
with my i1 Display on repeated white measurements [RMS delta E 0.021
versus 0.13 - nearly too good to be true...]

Regards,
Gerhard

Other related posts: