[Insight-users] Running multilevel image registration

[Luis Ibanez]

Hi Motes,

Thanks for the clarification, and for letting us know
that you solved the problem.

The increase in the number of samples certainly
will explain a longer computation time.

Computation time with MI should be roughly proportional
to the number of spatial samples.


     Thanks


          Luis



----------------------------------------------------------------------------
On Fri, Dec 4, 2009 at 2:37 AM, motes motes <mort.motes at gmail.com> wrote:
> On Thu, Dec 3, 2009 at 2:16 AM, Luis Ibanez <luis.ibanez at kitware.com> wrote:
>> Hi Motes,
>>
>>
>> 1) What Image metric are you using ?
>>
>
>
> Problem solved. I am using
>
>          typedef
> itk::AdvancedMeanSquaresImageToImageMetric<FixedImageType,MovingImageType
>>                             MetricAdvancedMeanSquaresType;
>
> with a random sampler. When the second level is run the number of
> samples is raised from 10000 to  50000 which of course explains why it
> takes longer time.
>
>
>          typedef
> itk::AdvancedMeanSquaresImageToImageMetric<FixedImageType,MovingImageType
>>                             MetricAdvancedMeanSquaresType;
>          int number_of_samples = 10000;
>          MetricAdvancedMeanSquaresType::Pointer
> metricAdvancedMeanSquares = MetricAdvancedMeanSquaresType::New();
>          metricAdvancedMeanSquares->SetUseNormalization(true);
>          metricAdvancedMeanSquares->SetImageSampler(random_sampler);
>          registration->SetMetric(metricAdvancedMeanSquares);
>
>
>
>> 2) If it is Mutual Information, could you please make
>>    sure that you are using the same number of samples
>>    in both scenarios of BSpline grid 12 * 12 * 12    ?
>>
>> 3) When it takes only 8 seconds,...
>>    does it still run 10 iterations ?
>>
>> 4) Have you compiling your code for Release ?
>>    (you shouldn't measure times with code
>>     compiled for Debug).
>>
>>
>>  Please let us know,
>>
>>
>>         Thanks
>>
>>
>>             Luis
>>
>>
>>
>> ------------------------------------------------------
>> On Wed, Dec 2, 2009 at 12:49 PM, motes motes <mort.motes at gmail.com> wrote:
>>> I run a multi-level image registration process using the
>>> BSplineDeformableTransform on a 3D image volume.
>>>
>>> In the first level I run with 6*6*6 nodes. In the second level I run
>>> with 8*8*8 nodes and in the last level I run with 12*12*12
>>>
>>>  Each level is run for 10 iterations (I know the is very low but I am
>>> currently in a testing phase).
>>>
>>> Now when it gets to the third level (12*12*12) the time used is around
>>> 40 seconds. But if I start with the 12*12*12 cube it only takes around
>>> 8 seconds.
>>>
>>> Any ideas on what might be the reason that the 12*12*12 level takes
>>> longer time when executed last? I have checked that data is not
>>> duplicated or allocated twice anywhere.
>>> _____________________________________
>>> Powered by www.kitware.com
>>>
>>> Visit other Kitware open-source projects at
>>> http://www.kitware.com/opensource/opensource.html
>>>
>>> Kitware offers ITK Training Courses, for more information visit:
>>> http://www.kitware.com/products/protraining.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
>>>
>>
>