[Insight-developers] image PixelType when using RegistrationMetricv4

Thu Jan 17 17:34:43 EST 2013

Hi,

Yes, the issue was with the use of vnl optimizers and vnl cost functions,
which are untemplated.

I haven't looked at my notes fully, but I do see a comment that we might try
having the vnl wrappers make a conversion from float to double when needed,
ala the cast filter Xiaoxiao suggested.

I can't find a gerrit patch for this, which is strange. The work was done
around June 2011. I'll look some more.

Most of the work is tedious, i.e. in changing the optimizers to templated
classes and the cascade of changes that requires. But I don't remember it
being too difficult otherwise, until the vnl code of course. Careful casting
to double to maximize efficiency at that point should be pretty
straightforward, it seems.

-M

> -----Original Message-----
> From: insight-developers-bounces at itk.org [mailto:insight-developers-
> bounces at itk.org] On Behalf Of Johnson, Hans J
> Sent: Thursday, January 17, 2013 11:36 AM
> To: brian avants; Nicholas Tustison
> Cc: Insight Developers
> Subject: Re: [Insight-developers] image PixelType when using
> RegistrationMetricv4
> 
> Brian,
> 
> I'd like to revisit this again.  I think it may be short-sighted to only
> use double precision in the optimizers, many cases can benefit
> significantly by using the less precise and much faster single precision
> floating point.
> 
> I've seen dramatic speed benefits from using single precision floating
> point rather than double precision.
> 
> Hans
> 
> ==========================================================
> =======
> Hans J. Johnson, Ph.D.
> Assistant Professor, Department of Psychiatry
> 
> Mailing Address:
> W274 GH                             Email: hans-johnson at uiowa.edu
> 200 Hawkins Drive                   Phone: (319) 353 8587
> The University of Iowa
> Iowa City, IA 52242
> 
> 
> 
> 
> 
> 
> -----Original Message-----
> From: Brian Avants <stnava at gmail.com>
> Date: Thursday, January 17, 2013 9:59 AM
> To: Nick Tustison <ntustison at gmail.com>
> Cc: ITK <insight-developers at itk.org>
> Subject: Re: [Insight-developers] image PixelType when
> using   RegistrationMetricv4
> 
> hi nick , xiaoxiao
> 
> it's the optimization framework that forced this - michael stauffer
> did a pass through rewriting the optimization framework with a
> template parameter but this was nixed in one of the developer
> meetings.
> 
> agree that it's not ideal --- however, upside is that solutions should
> be more stable/precise, to state the obvious.
> 
> b
> 
> brian
> 
> 
> 
> 
> On Thu, Jan 17, 2013 at 10:55 AM, Nicholas Tustison <ntustison at gmail.com>
> wrote:
> > Yeah, this is a legacy issue that we need to address but I don't know
how
> > easy it's going to be.  It's due to the fact that the transform
framework
> > requires
> > double type which predates our work.  Brian, Michael and I will have to
> > discuss this to see if we can fix it or at least minimize the places
> >where
> > double
> > is required.
> >
> > Nick
> >
> >
> > On Jan 17, 2013, at 10:50 AM, Xiaoxiao Liu <xiaoxiao.liu at kitware.com>
> >wrote:
> >
> > Hi Nick,
> >    With the recent changes in the RegistrationMetricv4, seems that the
> > PixeType of an image needs to be always set to double?
> >    Old registration code  that with float type fails to compile, but
runs
> > fine with double :
> >
> >https://github.com/InsightSoftwareConsortium/ITKv4-
> TheNextGeneration-Tutor
> >ial/commit/20df5ac66d43061b5d1ca3665dd1712a5986a879
> >    The internal parameter measurements are using double, but is there a
> > reason to require the image type to be double as well? Am I missing
> > something?
> >    Please give me some pointers. Thanks.
> >
> > -- Xiaoxiao
> >
> >
> > ---------------------------------------------
> > Xiaoxiao Liu, Ph.D.
> > R & D Engineer
> > Kitware Inc.
> > Clifton Park, NY
> > Phone: (518) 881-4924  or  (518) 371-3971 x124
> >
> >
> >
> > _______________________________________________
> > 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://kitware.com/products/protraining.php
> >
> > 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-developers
> >
> _______________________________________________
> 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://kitware.com/products/protraining.php
> 
> 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-developers
> 
> 
> 
> ________________________________
> Notice: This UI Health Care e-mail (including attachments) is covered by
the
> Electronic Communications Privacy Act, 18 U.S.C. 2510-2521, is
confidential
> and may be legally privileged.  If you are not the intended recipient, you
are
> hereby notified that any retention, dissemination, distribution, or
copying of
> this communication is strictly prohibited.  Please reply to the sender
that you
> have received the message in error, then delete it.  Thank you.
> ________________________________
> _______________________________________________
> 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://kitware.com/products/protraining.php
> 
> 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-developers