[Insight-users] OnePlusOne Optimizer - results are always 0

[Luis Ibanez]

Hi itkvtk123

Thanks for the update.

* Epsilon will define the convergence criterion
   So, it effects will not be visible at the beginning
   of the registration process.

You may want to play with the following parameters

* InitialRadius
* ShrinkFactor
* GrowthFactor

These parameters define the "sphere" around the
current parametric point, in which the optimizer
will try new values.

Note that this is done in the parametric space of
the transform, not in physical space.

The value of InitialRadius, relates to the settings
of the parameters scales.  For example, if you
double all parameter scales values, and you
double the InitialRadius, then the optimizer will
behave the same.

The rule of thum for setting parameter scales
is to set rotations to 1.0, so that you can assume
that the InitialRadius is set in units of radians
(for example : 0.1 radians, which is ~5.7 degrees).

Then you set the translation scales in order to make
that a small step of rotation is proportional to a
small scale of translation.

In your case:

  translationscale = 0.1 radian / ( 512 x 1.0 mm)


BTW: Is the final rotation of 25 degrees what you were
expecting ?  This rotation is close to the point of being
too large to be solved by registration. You may simply
need a better initialization of the Versor transform...


Regards,

       Luis


----------------------------------------------------
On Wed, Sep 9, 2009 at 9:42 AM, <itkvtk123 at gmx.net> wrote:

> Hi Luis.
>
> Thank you for your prompt answer.
> I found out that my scales cause the results of 0.
> When I comment the scales part, I get results which are different from 0,
> but which are terribly wrong.
> Epsilon does not seem to have much influence in the results, same with the
> initial transforms.
> (It is almost the same (wrong) result when I initialize the initial
> transforms to 0 or when I use the Initializer).
>
> In my test case it should detect a X movement of 5, Y movement of 5 and
> rotation of 5 degrees.
> But it actually detects X movement of 10.3, Y movement of -4.4 and rotation
> of -25.2 degrees.
> (By the way X is always wrong in sign regardless which optimizer I use)
>
>
> Is there a rule of thumb for parameters (Epsilon, scales, ...) for 512x512
> px (spacing 1mm) images (same modality) when I expect X and Y movement of +-
> 100 mm each and rotational movement of +- 5 degrees?
>
> Kind Regards,
> itkvtk123
>
>
>
> --
> Jetzt kostenlos herunterladen: Internet Explorer 8 und Mozilla Firefox 3 -
> sicherer, schneller und einfacher! http://portal.gmx.net/de/go/chbrowser
> _____________________________________
> Powered by www.kitware.com
>
> Visit other Kitware open-source projects at
> http://www.kitware.com/opensource/opensource.html
>
> Please keep messages on-topic and check the ITK FAQ at:
> http://www.itk.org/Wiki/ITK_FAQ
>
> Follow this link to subscribe/unsubscribe:
> http://www.itk.org/mailman/listinfo/insight-users
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20090909/f874c42d/attachment.htm>