[Insight-users] Execution time of the 4 steps of an image-registration iteration

[flo]

Hi,

I'm complety new to ITK so pardon some lack of knowledge about the
toolkit.
I want to test the possibilities of speeding up the image registration
process of the insight toolkit by porting parts to the GPU.

Has anybody some practical experience about execution time of image
registrations processes?
Surely it depends on the the fixed and moving image, the used
optimizer/metric/interpolation/transform and many other factors.
But it would help a lot if its possible to narrow the execution time
down to something like, the interpolation uses 50% of the time and the
transform takes just 5%.
Would be really some help to know which component is a good start that i
could start to optimize.

Otherwise has anybody a suggestion how to measure the execution time
more or less precisely in an efficient way?
Its really hard to understand the dependency between all used classes
and the whole execution happens in just one code-line
(registration->start) and therefor it is really hard to trace
back/debug.

I would be really thankful for some of your experiences about execution
times or maybe a suggestion how to do the measuring.

Florian Jung