[ITK-users] [ITK] Speed of ITKv4 based Gradient Descent Optimizer

[Andras Lasso]

See this long thread about this issue:
http://slicer-devel.65872.n3.nabble.com/BRAINS-rigid-registration-td4032229.html

Short summary: There were some performance regressions in mutual information metric computation, threading, etc. that have been recently fixed, so use the latest ITK version. ITKv4 uses a slightly slower but more accurate optimizer, but if you adjust the number of samples and stopping condition then usually the registration time only increases by about 30-50% (and you might get better quality results).

You can see all the changes that have been done in 3D Slicer and BRAINS repositories to go from a 10x-100x slowdown to an approximately equivalent registration speed with ITKv4 compared to ITKv3.

Andras

From: Community [mailto:community-bounces at itk.org] On Behalf Of Jiahui Wang
Sent: Monday, October 06, 2014 10:27 AM
To: insight-users at itk.org
Subject: [ITK] [ITK-users] Speed of ITKv4 based Gradient Descent Optimizer

Dear All,
I modified ImageRegistration8.cxx to run my own rigid registration.

However, I found that the processing time of ITKv4 based RegularStepGradientDescentOptimizer is almost 10 times more than that of ITKv3 based one.

The program was run on the same machine. The only difference is the ITK libraries.

Could anyone give me some suggestions to solve this issue?
Thanks,
JW

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://public.kitware.com/pipermail/insight-users/attachments/20141006/230f318a/attachment.html>