[ITK] Fwd: Image Pyramid Speed

[Stephen Parsons]

I believe we figured out the problem. We were setting the number of spatial
samples in the Mattes Mutual Information Metric to a large value (60% of
the pixels) and this value was not being changed in the levels of the image
pyramid. So while it was operating on images with fewer pixels at each
level, it was taking the same (massive) number of samples, which took just
as long.


On Tue, May 20, 2014 at 9:24 AM, Nicolas Gallego <nicgallego at gmail.com>wrote:

>
> Good morning Stephen,
>
> I have read recently de multi resolution pyramid filter and I just read
> the documentation for the registration method (version 4.5.0)
> there is a warning about the discrepancy of requested levels and a default
> schedule that one should check.
>
> It could be easier to point your issue if post the portion of the code you
> are testing.
>
> It may depend also of the particular parameters you are using
> or the way you are measuring time, see
>
> http://www.itk.org/Wiki/ITK/Examples/Utilities/TimeProbe
>
> this watcher can inspire you to check some events, progress and time
> measurements
> http://www.itk.org/Wiki/ITK/Examples/Utilities/SimpleFilterWatcher
>
> hope that helps
>
> --
> Nicolás Gallego-Ortiz
> Université catholique de Louvain, Belgium
>
>
>
> 2014-05-19 14:34 GMT+02:00 Stephen Parsons <stephen.parsons at uky.edu>:
>
>> I have been attempting to use the MultiResolutionImageRegistrationMethod.
>> As I understand it, the input images are blurred and then resampled to
>> create smaller internal images on which the registration operates for the
>> lower levels. I have set the number of levels and the starting shrink
>> factors and by outputting the shrink factor at each level I have confirmed
>> that it seems to be set up correctly (shrink factors are halved with each
>> new level).
>>
>> The program will run and produce correctly registered output, but each
>> iteration takes just as long as it did before I implemented the image
>> pyramid, when it was operating on full size images. Running smaller images
>> through the pipeline leads to a dramatic increase in speed which made me
>> think that the image pyramid in MultiResolutionImageRegistrationMethod
>> would allow for the same speed increase, at least in the beginning levels.
>> With the image pyramid there is no speed decrease as it proceeds through
>> the pyramid levels, the first takes just as long as the last.
>>
>> What could I have missed that would make the image pyramid just as slow
>> as not having it implemented at all? Have I misunderstood something about
>> the way it is implemented or is intended to be used?
>>
>> Thanks,
>> Stephen Parsons
>> University of Kentucky
>>
>>
>> _______________________________________________
>> Community mailing list
>> Community at itk.org
>> http://public.kitware.com/cgi-bin/mailman/listinfo/community
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/community/attachments/20140520/005cd26f/attachment-0002.html>