[Insight-developers] Re: ImageAdaptors, PixelAccessors and Neighborhood iterators

[Joshua Cates]

Hi Luis,

I agree it would be nice to support adaptors with the neighborhood
filters.  But we should be very careful since these are significant
changes to code used by a large percentage of the toolkit filters
(nonlinear diffusion, level-sets, convolution, some deformable
registration, and assorted image processing).

I propose we first run extensive benchmarks (3D, large volumes, multiple
platforms) on all the relevant code to make sure adaptor support does not
introduce an unacceptable performance penalty.  It would be nice to have
these baselines anyway. If the increased complexity is optimized out in
the common case, then we may be ok.

This would be a good topic for the TCON.  I'd like to hear other
developers opinions on these sorts of speed/convenience tradeoffs, i.e.  
how much performance hit is acceptable to folks in cases like this.

Josh.

______________________________
 Josh Cates			
 Scientific Computing and Imaging Institute
 University of Utah
 Email: cates at sci . utah . edu
 Phone: (801) 587-7697
 URL:   http://www . sci . utah . edu/~cates


On Tue, 28 Oct 2003, Luis Ibanez wrote:

> 
> Hi Josh,
> 
> I'm tracking down a problem reported in the users-list
> concerning the use of the Laplacian filter over an image
> connected to an ImageAdaptor.
> 
> The bottom line seems to be that the Neighborhood iterators
> do not support the use of ImageAdaptos and PixelAccessors.
> 
> I understand that some performance issues may be involved,
> nonetheless, I gave it a shoot and got a set of modification
> that allow me to compile the Laplacian filter with an adaptor.
> 
> Before going too far I would like to get your opinion on
> how invasive this could be, and how bad could it affect
> performance.
> 
> The CVS diffs for these changes are at the end of this email.
> 
> Please let me know what you think about these potential
> changes.
> 
> 
> Thanks
> 
> 
>    Luis
> 
> 
> ---------------------
> 
>