[Tubetk-developers] itkImageToTubeRigidRegistration2

[Matt McCormick]

Hi Michael,

Many thanks for the information.


>
> From what I remember and what I just checked quickly:
>
> The "itkImageToTubeRigidRegistration" is the re-factored and cleaned version
> of the initial one given the exact same results while the
> "itkImageToTubeRigidRegistration2" is a modified version of this previous
> class.
> This last one is supposed to be better as described here:
> (https://docs.google.com/open?id=1psAY1BKx2pEfVPDF4S-xwHOuSP4Mrp7AxQXErgmmQvHthA-LvFauCEdfpQQs).

Could you please share that document with me?  The link does not work.

> It is the one I would like to be the right one (I may have to go through an
> other clean step however), nevertheless I was waiting for some feedbacks
> first and as you are saying:
>
>> Comparing the output of the tests, the registered tube resampled onto
>> an image, the "2" result does not resemble the input image (as far as
>> I can tell).
>
> --> See Image enclosed.
>
> I just have some trouble to explain this as the performance measurement on
> the second metric seems to be more accurate and following a "more" coherent
> curve and equations compared to the initial one (c.f. previous link &&
> enclosed convergence curves). I was thinking about first making the read of
> the ".tre" file to be able to make some visual comparisons; I am really open
> to any explanations/ideas about why the results are that "bad" given the new
> metric.

I have attached screenshots from the test outputs on my side.  The
translation along the "axial" direction is off in the "2" case.

If you would like a second set of eyes, I can work on a piecewise
integration of the second class.

>
> Then, concerning the ability to read a ".tre" file into a viewer, It is
> exactly what I am currently working on. I will create a specific module to
> get the reading and representation done as well as the integration with a
> MRML scene.
> Thank you to let me know if the plan suits you and do not hesitate for any
> comments/suggestions.

Excellent.  Where is this code currently located?  Would you like to
work on this while I address the convergence issues?

Thanks,
Matt



>
> Regards,
> Michael J-L.
>
>
>
> On Tue, Jul 24, 2012 at 3:28 PM, Danielle Pace <danielle.pace at kitware.com>
> wrote:
>>
>> Unfortunately, there's no way to read a .tre into Slicer - that's
>> something we've always wanted to do.
>>
>> Stephen has a program to visualize .tre's - perhaps he can copy it for
>> you, or perhaps there's a more generic ITK spatial object viewer.
>>
>> -Danielle
>>
>>
>> On Tue, Jul 24, 2012 at 2:11 PM, Matt McCormick
>> <matt.mccormick at kitware.com> wrote:
>>>
>>> On Tue, Jul 24, 2012 at 5:44 PM, Stephen Aylward
>>> <stephen.aylward at kitware.com> wrote:
>>> > Michael extended his vacation by one day.   He should be back in the
>>> > office tomorrow.
>>>
>>> OK, great.
>>>
>>> >
>>> > Thanks for looking at this.   My understanding is that the 2 classes
>>> > are the correct ones; however, I really hate the "2" suffix.   If the
>>> > old ones were broken, they should be replaced - instead of using a
>>> > numbering system.
>>> >
>>> > Perhaps beging by looking for the differences in the code and see if
>>> > the "" versions can be replaced by the "2" version (nearly the same
>>> > APIs, etc).
>>>
>>> As far as I can tell, the Registration classes are nearly the same
>>> apart from some cleanup.  The Metric class underwent refactoring.  The
>>> best hint as to what is going on appears to be:
>>>
>>>  * The improvment implemented here is about
>>>  * the precomputation of the internal kernels proposed by Lange et al.:
>>>  * \link http://www.zib.de/lamecker/publications/cars2007.pdf
>>>
>>> Comparing the output of the tests, the registered tube resampled onto
>>> an image, the "2" result does not resemble the input image (as far as
>>> I can tell).
>>>
>>> Is there a way to visualize the tube spatial object directly?  It is a
>>> MetaIO ".tre" file containing a Scene, Group, and Tube ObjectType.  I
>>> assume Slicer should recognize it as a "Model", but it seems to think
>>> it is a "Volume".
>>>
>>> Thanks,
>>> Matt
>>>
>>>
>>> >
>>> > s
>>> >
>>> > On Tue, Jul 24, 2012 at 1:04 PM, Matt McCormick
>>> > <matt.mccormick at kitware.com> wrote:
>>> >> The test output of the "2" classes does not appear to be correct?
>>> >>
>>> >> On Tue, Jul 24, 2012 at 4:18 PM, Matt McCormick
>>> >> <matt.mccormick at kitware.com> wrote:
>>> >>> Hi Michael,
>>> >>>
>>> >>> I am looking at the TubeTK itkImageToTubeRigidRegistration,
>>> >>> itkImageToTubeRigidRegistration2, itkImageToTubeRigidMetric,
>>> >>> itkImageToTubeRigidMetric2, and I am wondering if we replace the
>>> >>> original classes with the "2" version, and concentrate on development
>>> >>> there?
>>> >>>
>>> >>> Thanks,
>>> >>> Matt
>>> >> _______________________________________________
>>> >> Tubetk-developers mailing list
>>> >> Tubetk-developers at tubetk.org
>>> >> http://public.kitware.com/cgi-bin/mailman/listinfo/tubetk-developers
>>> >
>>> >
>>> >
>>> > --
>>> >
>>> > ==============================
>>> > Stephen R. Aylward, Ph.D.
>>> > Senior Director of Operations - North Carolina
>>> > Kitware, Inc. - North Carolina Office
>>> > http://www.kitware.com
>>> > stephen.aylward (Skype)
>>> > (919) 969-6990 x300
>>> _______________________________________________
>>> Tubetk-developers mailing list
>>> Tubetk-developers at tubetk.org
>>> http://public.kitware.com/cgi-bin/mailman/listinfo/tubetk-developers
>>
>>
>>
>>
>> --
>> Danielle Pace, M.ESc.
>> Research and Development Engineer
>> Kitware Inc.,
>> North Carolina Office
>>
>> www.kitware.com
>> 919-969-6990 X 319
>>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: itkImageToTubeRigidRegistrationTest2Output.png
Type: image/png
Size: 58022 bytes
Desc: not available
URL: <http://public.kitware.com/pipermail/tubetk-developers/attachments/20120725/b53b03c8/attachment-0004.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: itkImageToTubeRigidRegistrationTestOutput.png
Type: image/png
Size: 88306 bytes
Desc: not available
URL: <http://public.kitware.com/pipermail/tubetk-developers/attachments/20120725/b53b03c8/attachment-0005.png>