[Insight-users] MPI Vector Gradient Anisotropic Diffusion Image Filter

[Bradley Lowekamp]

Thanks for sharing that! I was just squeezing in my images into main memory before, and now they are going to be  getting 16x bigger per slice, so I needing a very similar solution! This is great!

Earlier I was looking at the scalability of ScalarAnisotropic diffusion. And found that the single threaded CalculatedAverageGradienMagnitudeSquared began dominating computation time. So I began work on multi-threading the method. You can see the work in progress here:
https://github.com/blowekamp/ITK/commit/ae4103b58098a52380930c1812d27c0fbe9f9f25

I should return to that work to finish it up.

I hope you make this into an InsightJournal contribution.

Brad




On Dec 7, 2010, at 1:28 PM, Kevin H. Hobbs wrote:

> On 12/06/2010 02:25 PM, Bradley Lowekamp wrote:
>> Hello Kevin,
>> 
> If you changed the
>> output image to a image of vector is should go away. 
>> 
> 
> Yup this gets rid of the warning
> 
>> BTW: I am curious of you are communicating between each node after each
>> iteration, or are just having a large overlapping region?
> 
> I changed the code to only pad each region by 1 pixel in each direction
> and communicate the boundary pixel values at each iteration. The runtime
> dropped to 19 minutes, and the ram used was a little less. It should
> make the number of iterations unlimited ( no more excess padding soaking
> up RAM ).
> 
> I'll have to run a test on a small image to see if everything works when
> compared to regular Vector Gradient Anisotropic Diffusion.
> <MPIVectorGradientAnisotropicDiffusionImageFilter.cxx>

========================================================
Bradley Lowekamp  
Lockheed Martin Contractor for
Office of High Performance Computing and Communications
National Library of Medicine 
blowekamp at mail.nih.gov


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.itk.org/pipermail/insight-users/attachments/20101207/a8988c75/attachment.htm>