[Insight-users] ICP progress monitoring still not working(?)

[Luis Ibanez]

Hi Patrick,

You are right, the ITK adapters to MutlipleValued VNL optimizers
were not yet reporting iteration events during the optimization.

We just committed a fix for this lack of support. Mainly copying it
from the Single Valued optimizers, that already have it implemented.

You may have to update you CVS checkout of ITK in order to get access
to this functionality.


About your number of iteration settings:

Using 10 iterations for ICP is extremely optimistic   :-)
It is not surprising that the optimizer exceeds that small number
of iterations and returns to you telling you that the optimizer
stops because it has performed already too many iterations for
the limit that you set.

Please be more generous with the number of iterations, you may
want to consider something in the range from 100 to 500.

The discrepancy between the number of iterations set at the ITK
level and the ones reported by VNL are due to the which netlib
library does VNL invokes for solving the optimization. In one
case (lmder) it is considered that the number of function evaluations
is the one that corresponds to the user setting for number of
iterations.

Note that the message is coming from the file:


     Insight/Utilities/vxl/core/vnl/algo/
                  vnl_levenberg_marquardt.cxx


You may want to post this question to the vnl-users mailing list.



Please let us know if you have further questions,


    Thanks


       Luis



======================
Patrick De Smet wrote:
> 
> Dear all,
> 
> according to some initial experiments I did, quite some doc digging, and 
> finally
> http://public.kitware.com/pipermail/insight-users/2006-February/016612.html
> it would seem it is not possible to monitor ICP. Is this still true?
> 
> Are there going to be any future changes to this situation?
> 
> Do you have suggestions for another fast solution
> ("fast": adaptation/easy to understand  ICP-source code package) that I 
> could use instead?
> 
> Also, if I set  numberOfIterations =  10;
> in IterativeClosestPoint2.cxx
> with some simple experimental point sets I get
> 
> vnl_levenberg_marquardt: too many iterations
> vnl_levenberg_marquardt: 2 iterations, 10 evaluations, 8 residuals.  RMS 
> error start/end 1.0247/0.560419
> 
> Ok, it is a VNL problem not ITK (?)  but maybe someone could crosspost 
> to point out
> that their errors are confusing/not explaining a lot : 2 iterations <> 
> 10 --> error should read "too many evaluations" ?
> 
> Thanks!
> regards,
> P.De Smet
> 
> 
> _______________________________________________
> Insight-users mailing list
> Insight-users at itk.org
> http://www.itk.org/mailman/listinfo/insight-users
> 
>