[argyllcms] Re: Dispcal - Inverting Jacobian failed?
- From: Gerhard Fuernkranz <nospam456@xxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 22 May 2009 11:27:53 +0200
Graeme Gill wrote:
> That model is the Jacobian matrix. Normally it updates the Jacobian
> with each measurement attempt, but if the Jacobian prediction proves
> too inaccurate, it re-measures the behavior of the display at that
> point in order to re-establish the Jacobian. It measures the
> forward behavior for a small change in the RGB values, and then
> inverts the resulting matrix to form the Jacobian.
I'm still surprised. Even if the device response is flat in this region,
I guess it still not so likely that this immediately leads to a singular
Jacobian matrix, due to noise in the XYZ readings. IMO it requires quite
some luck to obtain a random matrix which is not invertible. So I'm
wondering whether there might be something wrong with the instrument
readings as well (e.g. clipping or too coarse quantization, so that no
noise is returned). It may be worth to look at the actual internal
numbers and check if they are reasonable.
Alastair, is this problem reproducible, or sporadic?
Which of the three error messages did you get?
* "Inverting Jacobian failed (1)" or
* "Inverting Jacobian failed (2)" or
* "Inverting Jacobian failed (3)"
[ I suspect, it was (3)? ]
Regards,
Gerhard
Other related posts:
