[ITK] rewrite the registration metric using CUDA

Matt McCormick matt.mccormick at kitware.com
Sun Jan 24 14:55:28 EST 2016


Hi Wu,

Someone that has gone down this route already may have some additional
input. Metric computation is a major source of computation during
registration.  But, other parts of the analysis may need to be ported
to the GPU to get any actual speed-ups -- transferring data to and
from CPU / GPU can be very expensive.

HTH,
Matt

On Sun, Jan 24, 2016 at 2:39 PM, wu qiu <qiu.wu.ch at gmail.com> wrote:
> Hi, Matt,
>  Thanks for your response.
>  Basically, what I am trying to do is to speed up rigid or affine
> registration. After investigating ITK registration source code, I found that
> there are not too much to do with optimization procedure, or say, too
> difficult for me to improve the optimization part.  Speeding up the
> calculation of the similarity metric could be a good choice for me. I took a
> look at the filter of itkMeanSquaresImageToImageMetric, and found it is
> already multi-thread coded. Do you think it is worth of rewriting it in
> CUDA? I don't think I am the first one to come up with this question. Anyone
> who did this job could mind to comment?
>
> Cheers,
>
> Wu Qiu, Ph.D
> Calgary Stroke program,
> the University of Calgary,
> Email: qiu.wu.ch at gmail.com
>
> 2016-01-23 20:30 GMT-07:00 Matt McCormick <matt.mccormick at kitware.com>:
>>
>> Hi Wu,
>>
>> This contribution would be very welcome! There are examples here [1]
>> [2] [3] that could be consolidated with your registration code into an
>> ITK module [4].
>>
>> Thanks,
>> Matt
>>
>> [1] https://github.com/SimonRit/RTK
>>
>> [2] http://www.insight-journal.org/browse/publication/802
>>
>> [3] http://www.insight-journal.org/browse/publication/803
>>
>> [4]
>> http://itk.org/ITKSoftwareGuide/html/Book1/ITKSoftwareGuide-Book1ch9.html#x50-1430009
>>
>> On Sat, Jan 23, 2016 at 10:23 PM, wu qiu <qiu.wu.ch at gmail.com> wrote:
>> > Hi,
>> > I want to speed up or rewrite the calculation of registration metric
>> > used in
>> > the rigid and affine registration, such as
>> > itkMeanSquaresImageToImageMetric,
>> > using CUDA.
>> > Anyone who has experiences could give me some clues or some examples I
>> > can
>> > learn? Thanks.
>> >
>> > Cheers,
>> >
>> > Wu
>> >
>> >
>> >
>> > _______________________________________________
>> > Community mailing list
>> > Community at itk.org
>> > http://public.kitware.com/mailman/listinfo/community
>> >
>
>


More information about the Community mailing list