[Insight-users] Testing GPU-based Median Filters

[Richard Beare]

Hi,

Can you provide any examples of how the factory mechanism works with
filter classes? I'm familiar with the IO factories but haven't dealt
with anything like factories for standard filter classes.

Also, as far as kernel size goes, I suspect large sizes aren't seen as
useful because most implementations get too slow as the size grows.
Having access to versions that don't depend on kernel size can open up
many new options.

e.g. very large (111x111) robust smoother based on a median filter for
background intensity estimation in 2kx2k microscopy images. A
histogram based, separable kernel can run fast (~2 seconds). The
alternative is to mess around with resampling etc. Note that the
separability is an approximation, so this is a smoother, not a 2D
median operation.

On Wed, Jan 13, 2010 at 6:36 AM, Bill Lorensen <bill.lorensen at gmail.com> wrote:
> In fact, if the API is the same as the itk filters, you might be able
> to use itk's factory mechanism to load your accelerated classes at
> run-time.
>
> Bill
>
> 2010/1/12 Gaëtan Lehmann <gaetan.lehmann at jouy.inra.fr>:
>>
>> Le 12 janv. 10 à 00:18, Christopher Share a écrit :
>>
>>> Hi,
>>>
>>> I'm in the process of developing several different GPU-based median
>>> filters for the ITK (as part of the 09/10 VPAC Summer Project - this year's
>>> project doesn't have a web page yet but here's a link to last years' -
>>> http://www.vpac.org/?q=node/290). I have a question regarding the testing of
>>> the filters. I'd like to test the performance of the GPU filters under
>>> different conditions, however I'm not clear about how these filters are used
>>> in practice.
>>>
>>> For the purposes of testing the filters I have the following questions:
>>>
>>> 1. What image sizes should be tested? I was thinking of "small", "medium"
>>> and "large" but I'm not sure what these sizes would be wrt to the ITK.
>>
>> I think the test should cover images smaller than the GPU memory and larger
>> than the GPU memory.
>>
>>>
>>> 2. Should I test RGB/RGBA images as well as greyscale images?
>>
>> no: ITK's filters does not support them.
>>
>>>
>>> 3. For the median filter case, what radius's should be tested?
>>
>> You can have a look at ITK's test for that filter. I think that reusing the
>> same test with your filter is a good start.
>>
>>> Should different radius's in different directions be tested?
>>
>>
>> Yes, I think it should, because ITK's filter support it.
>>
>> BTW, it would be great if the filter can have the exact same API than
>> MedianImageFilter. That way, it can be used as a direct replacement.
>>
>> Gaëtan
>>
>> --
>> Gaëtan Lehmann
>> Biologie du Développement et de la Reproduction
>> INRA de Jouy-en-Josas (France)
>> tel: +33 1 34 65 29 66    fax: 01 34 65 29 09
>> http://voxel.jouy.inra.fr  http://www.itk.org
>> http://www.mandriva.org  http://www.bepo.fr
>>
>>
>> _____________________________________
>> 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
>>
>>
> _____________________________________
> 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
>