[ITK] Metric Evaluation

[Yaniv, Ziv Rafael (NIH/NLM/LHC) [C]]

Hello Wassim,

The following SimpleITK Jupyter notebook may be of interest to you (uses the exhaustive optimizer as an initialization approach): https://github.com/InsightSoftwareConsortium/SimpleITK-Notebooks/blob/master/63_Registration_Initialization.ipynb.

Based on your runtime of 5sec I suspect you are not using sampling when computing the metric. Possibly consider random or uniform sampling. There are cases where you can achieve a decent registration using less than 10% of the data.

       hope this helps
              Ziv


From: wassim bouzeid <abou_zeid_wassim at hotmail.com<mailto:abou_zeid_wassim at hotmail.com>>
Date: Wednesday, May 25, 2016 at 10:05 AM
To: "community at itk.org<mailto:community at itk.org>" <community at itk.org<mailto:community at itk.org>>
Subject: [ITK] Metric Evaluation

Hi Everyone,

I want to explore the domain of an image metric by exhaustive search (moving the second image with respect to the first image from a negative position to a positive position) so I am using the example in the source code MeanSquaresImageMetric1.cxx on 2D images. The time to compute the metric value for a specified parameters on these 2D images is approximately 0.02 sec. (PC with 32000 MB memory, and 1 cores @ 3201 MHz).

When I am applying this example to 3D nifti images, the computation time of the metric value is large (more than 5 sec.).

Does anyone know how to speed up the computation time on 3D images without using an optimizer?

Thanks in advance.

Best,
Sam

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/community/attachments/20160525/7c8b32d9/attachment.html>