[Insight-users] MPI Fast Marching

[Luis Ibanez]

Hi Kevin,


1) I don't think MPI is available in the testing environment of the
    Insight Journal.

    I'm not sure how much effort would be involved in setting
    up MPI and maintaining it. Could you please log a feature
    request in the Insight Journal ? :

        http://public.kitware.com/Bug/my_view_page.php

    (Select "Insight Journal" in the Project : drop down menu
     on the upper right).


2)  YES, OPEN ACCESS  Contributions should be rated
     Higher by Universities and Funding agencies than
     papers published if journals that restrict access and
     enforce restrictive copyrights. Traditional journals
     restrict the dissemination of scientific information and
     stifle progress. Public funds shouldn't be used for paying
     Journals that restrict the dissemination of information.

     If your grants are from NIH, you are actually required
     by Law to make your publications available, under the
     NIH Public Access Policy:

             http://publicaccess.nih.gov/policy.htm



3) You approach for managing tolerances sounds reasonable.
    It seems to be similar to the progressive reduction of
    Step Length performed by the RegularStepGradient
    descent optimizer.

4) Would you have any objections for us creating a
    directory for you in the NAMIC Sandbox:
      http://svn.na-mic.org/NAMICSandBox/trunk/

    In this way, other developers will be able to look at
    the code and giving it a try even before you submit it
    to the Insight Journal.


Please let us know,


      Thanks


             Luis


-------------------------------------------------------------
On Fri, Oct 23, 2009 at 11:54 AM, Kevin H. Hobbs <hobbsk at ohiou.edu> wrote:
> On Tue, 2009-10-20 at 12:10 -0400, Luis Ibanez wrote:
>> Hi Kevin,
>>
>> This is great !
>>
>> I hope you are planning on submitting it to the Insight Journal.
>
> Is MPI available in the IJ environment?
>
> It is my desire to submit it to the insight journal.
>
> Realizing my desire during work hours however is up to my boss and
> whether or not the lab gets any credit from funding agencies or Ohio
> University come grant time.
>
> Sounds like Open Access Week Day #3 policy 1 to me.
>
>>
>>
>> Regarding your point about the float comparison, I agree with
>> you in that an exact comparison is unrealistic.  You may want
>> to consider the use of the following "epsilon" values:
>>
>>               vcl_numeric_limits<double>::epsilon()
>>               vcl_numeric_limits<float>::epsilon()
>>
>> as tolerances for the comparison.
>>
>>
>> Another option is to use the value:
>>
>>             itk::NumericTraits< float >::min()
>>
>
> Thank you for these comments.
>
> These numbers are tiny.
>
> I am thinking about the re-setting of the trial and alive points between
> iterations.
>
> I can easily see the level plus or minus a tolerance falling completely
> between the values at each point.
>
> A 1d ASCII art speed image :
>
> .5 .5 1 1 1 1.5 1.5 2 2
>
> split into two padded images :
>
> .5 .5 1 1 1
>          1 1.5 1.5 2 2
>
> fast marching seeded at both ends with value 0 :
>
> 0  2  3 4 5
>          2.8 1.8 1.2 .5 0
>
> Now both pieces need to recompute their trial and alive points based on
> the minimum value of their boundaries after MPI communication. For the
> left piece 2.8 plus or minus a tiny value will not equal any value in
> the left piece, so I made a neighborhood iterator walk the image and set
> the points as; alive if its value and all its neighbor's values are less
> than 2.8, trial if its value is less than 2.8 and any of its neighbors
> values are greater than 2.8, and nothing was set if the value was above
> 2.8.
>
> A  T  N N T
>          A   A   A   A  A
>
> The next round of fast marching would produce something like :
>
> 0  2  3 3.8 2.8
>            2.8 1.8 1.2 .5 0
>
> Does this seem like the right way to go about it?
>
> I've attached my most recent code with huge chunks of wrong stuff
> commented out.
>
>