Conclusion
n
With relatively simple port to CUDA slow
iterative procedure can run at interactive speed
n
Just two kernels and few optimization steps
n
Possible CUDA optimization of the entire image
processing pipeline for the eye surgery
SIAM MI09