[Insight-users] FW: Image Registration examples test results

[Xu, Yanjun]

Hi, Luis,

Thanks for the explanation. That really helps. Most differences are very
small except the ImageRegistration5.exe (the rotation and translation
example). The rotation estimation has a pretty big error. And the iteration
number is much bigger than the expected number. After setting the
MaximumStepLength to 1.0 according to the guide, it gave the exactly correct
result. It seems that MaximumStepLength should be added in as an argument to
ImageRegistration5.exe. 

By the way, I couldn't run ctest. When I run ctest, it gave the information
that no test was found. I had the buiding test on in Cmake though. Do I have
to setup Dart to do the testing? 

Thanks a lot.

Sincerely,
Yanjun


-----Original Message-----
From: Luis Ibanez [mailto:luis . ibanez at kitware . com] 
Sent: Thursday, May 22, 2003 1:31 PM
To: Xu, Yanjun
Cc: 'Insight-users at public . kitware . com'
Subject: Re: [Insight-users] FW: Image Registration examples test results


Hi Xu,

You may expect all the multi-modality registrations to give slightly
different results due to the use of random sampling during the
evaluation of the mutual information metric. Apart from those, the
results should be pretty much the same as in the software guide.

Are the differences that you found larger than what you may expect
from numerical instability ?

Note that the examples for the software guide were all executed on
Linux. If you try this same code under windows you my find different
results due to the particular encoding of floating point numbers.
(40bits as opposed to the IEEE standard encoding). We have also found
that the numerical results are different if you run in Debug mode or
Release mode.

So...
a bit of numerical differences are to be expected in the results.

If you find large differences, that may indicate a bug in the code.

However, at this point the examples are verified nightly using a
regression testing system that you can check on the Dashboard.

Look for example at:
http://www . itk . org/Testing/Dashboard/20030522-0500-Nightly/TestDetail/__Exam
ples_Registration_ImageRegistration1Test.html

When any of those tests fail, a difference image is generated
to help in the identification of the problem.


Please let us know if the differences that you encounter are
unreasonable large.


--------------

BTW in order to run tests whithout having to worry about passing
the parameters you can use the command line utility "ctest".
Just move to the binary directory of the examples and type:

                 ctest

you can also do

                 ctest ImageRegistration2Test

of even do regular expressions with

                 ctest -R Image*

you can see what test would be run, if you type

                 ctest -N

a short help about the ctest option can be displayed
with
                 ctest -h


--

   Regards


      Luis