This is a quick and dirty benchmark of antsRegistration for memory and walltime usage by registering two high-resolution T1 brain images with varying levels of resampling.
All benchmarking was done in 8-CPU limited SGE parallel environment runs with 32 GB of ram made available to each job.
To run the benchmark:
#First you need two high resolution brains, named brain1_0.3.mnc and brain2_0.3.mnc
#Some are available here http://cobralab.ca/atlases/
#First script with create all the resampled brains, then use qbatch to submit jobs
> ./do_benchmark.sh
#Wait for all registration jobs to finish
> ./collect_stats.sh > stats.csv
> Rscript benchmarking.R
This benchmark requires minc-toolkit, ANTs (with MINC support) and R with the following R packages: ggplot2,doBy,rgl,scatterplot3d,minpack.lm,akima
Figures showing the memory and time usage as a funtion of fixed and moving resolution.
Non-linear fits of memory and walltime for use in registration prediction for clusters.
You can find results in Rplots.pdf