ITK/Release 4/Enhancing Image Registration Framework/Tcon 2010-11-03

Discussion items

Numerics - wrap everything so it's "easy" to switch something else in, e.g. eigen. the current framework somewhat follows this style but it's not consistent everywhere. e.g. direct use of vnl svd.

Numerics - some places in the toolkit explicitly invert matrices rather than using linear solvers from vnl and/or itpack.

Numerics - itpack appears to still live inside the FEM part of ITK and could use some additional wrappers to make the tools more widely available to the rest of the toolkit.

Registration - we should benchmark itk v3.20 and itkv4 as it is today.

Registration - we need a public location (on midas or nitrc) where we systematically assemble the data and publish registration benchmark results e.g. using Gang Song's system.

Registration - we want to expose the voxel-wise value of the metric derivative s.t. we can have a dense field for use in dense registration.

Registration - we need to allow symmetric / unbiased registration which will necessitate a parallel metric hierarchy or some relatively small changes to the current hierarchy.

Registration - Brad discovered a bug which he will send to the list and record in Mantis.

Registration - will meet sunday night in iowa to get started on refactoring.