[Insight-users] how to reduce the run time of ITK
Luis Ibanez
luis.ibanez at kitware.com
Wed Dec 29 08:26:04 EST 2004
Hi Yixun,
1) Yes, Multi-resolution will be helpful.
There are two types of multiresolution when you are dealing
with BSplines.
- Type A: use several resolutions on the BSpline grid
- Type B: use several resolutions on the image itself
and, you can combine both types.
You will find an example on how to use "Type A" multi-resolution
for BSplines in the file:
Insight/Examples/Registration/
DeformableRegistration6.cxx
The use of "Type B" multi-resolution for BSplines is
equivalent to what is illustrated in any of the other
multi-resolutions examples:
Insight/Examples/Registration/
MultiResImageRegistration1.cxx
MultiResImageRegistration2.cxx
2) ITK supports parallel computation on a case-by-case
basis. The reason is that not all algorithms are
suitable for parallelization.
If you are interested in the efforts on parallel computation,
you may want to join other ITK users with similar interest.
As a way to get started, we have created a page for a Working
Group on parallel computation in the Wiki pages:
http://www.itk.org/Wiki/ITK_Working_Group_Parallel_Computation
Regards,
Luis
------------------
Yixun Liu wrote:
> Thank for your reply! Will the multiResolution registration be
> helpful? is the example about
multiresolution regestration based on the BSpline available? Does ITK
support Parallel computation?
>
> Regards,
>
> Yixun Liu
>
> ----- Original Message -----
> From: "Luis Ibanez" <luis.ibanez at kitware.com>
> To: "Yixun Liu" <yxliu at fudan.edu.cn>
> Cc: <Insight-users at itk.org>
> Sent: Wednesday, December 29, 2004 6:27 AM
> Subject: Re: [Insight-users] how to reduce the run time of ITK
>
>
>
>>Hi Yixun,
>>
>>Thanks for posting the additional information about
>>the deformable registration that you are performing.
>>
>>
>>1) 40 minutes for a 3D deformable registration is
>> not unusual
>>
>>
>>2) Debug version is a bad choice. Please recompile
>> your code for Release. That will reduce the
>> computation time. Note that in Windows, you will
>> have to first recompile ITK for Release, and then
>> recompile your own application for release.
>>
>>
>>3) BSplines performance may be improved in the
>> future, (some months from now) new classes must
>> be added to the toolkit in order to implement a
>> faster evaluation of BSplines.
>>
>>
>>
>> Regards,
>>
>>
>> Luis
>>
>>
>>
>>------------------------
>>Yixun Liu wrote:
>>
>>
>>>Hi,
>>> 1) The operating system is winXP and compiler is VC++6.0
>>>2) I use the non-rigid registration. The metric is MattesMI and the transform is Cubic BSpline
>>>3) The size of the two images is: 256*256*46, spacing is 0.859*0.859*3
>>>4) the control grid is: 15*15*6 (include the control points outside the image)
>>>4)need about 40 minutes to finish
>>>5) the debug version
>>>7)hardware profile: intel IIII 2.6G CPU, 1.5G Memory.
>>>6) optimizer parameters are
>>>
>>>
>>>> optimizer->SetGradientConvergenceTolerance( 0.05 );
>>>> optimizer->SetLineSearchAccuracy( 0.9 );
>>>> optimizer->SetDefaultStepLength( 1.5 );
>>>> optimizer->SetMaximumNumberOfFunctionEvaluations( 1000 );
>>>
>>>
>>>Thanks!
>>>
>>>
>>>----- Original Message -----
>>>From: "Luis Ibanez" <luis.ibanez at kitware.com>
>>>To: "Yixun Liu" <yxliu at fudan.edu.cn>
>>>Cc: <insight-users at itk.org>
>>>Sent: Monday, December 27, 2004 9:58 PM
>>>Subject: Re: [Insight-users] how to reduce the run time of ITK
>>>
>>>
>>>
>>>
>>>>Hi Yixun,
>>>>
>>>>
>>>>We will need you to provide the following information,
>>>>in order to be able to give you any useful advice:
>>>>
>>>>1) What operating system and compiler are you using ?
>>>>2) What specific registration algorithm did you used ?
>>>>3) What time did it take you to register the images ?
>>>>4) What is the size (in pixels) of your images ?
>>>>
>>>>
>>>>The first thing that you should try in general is to select
>>>>
>>>>
>>>> CMAKE_BUILT_TYPE = Release
>>>>
>>>>
>>>>both when you configure ITK and when you configure your
>>>>application.
>>>>
>>>>
>>>>Please let us know about the registration method that you
>>>>used, since the options for speeding up are method-dependent.
>>>>
>>>>
>>>> Thanks
>>>>
>>>>
>>>> Luis
>>>>
>>>>
>>>>
>>>>-----------------
>>>>Yixun Liu wrote:
>>>>
>>>>
>>>>
>>>>
>>>>>Hi,
>>>>>
>>>>>I use the ITK to registe two 3D images. The registed result is fine, but
>>>>>the run time it too long. The registration parameters are finely
>>>>>adjusted. Does ITK support parallel computation? Please give me some
>>>>>suggestion about how to reduce the run time.
>>>>>
>>>>>Regards,
>>>>>
>>>>>Yixun Liu
>>>>>
>>>>>
>>>>>------------------------------------------------------------------------
>>>>>
>>>>>_______________________________________________
>>>>>Insight-users mailing list
>>>>>Insight-users at itk.org
>>>>>http://www.itk.org/mailman/listinfo/insight-users
>>>>
>>>>
>>>>
>>>>
>>>
>>
>>
>>
>
>
More information about the Insight-users
mailing list