Summer2009:Using ITK in python

• 

  Project Week Main Page

• 

  ITK provides body of algorithms in C++

• 

  Python provides interactive development environment.

Key Investigators

• GE: Jim Miller
• Isomics: Steve Pieper
• INRIA: Demian Wassermann

Objective

Develop strategies for embedding ITK filters in python code. See if this would be a workable strategy for creating slicer modules.

Approach, Plan

• Look at how numpy arrays are passed to C/C++ code using techniques like cython, weave, etc.
• Figure out what is needed to compile ITK code in these environments.
• Compare and contrast this approach to WrapITK

Progress

• Some pre-project week progress is shown on this page in the slicer.org wiki
• Debugging of python internals with Demian and discussion of future directions.
• Basic possibilities (all are somewhat convoluted):
  • WrapITK - combinatorial explosion; large libraries; C++ no longer available
  • Cython - yet another language; not clear how to use C++ idioms
  • Ctypeslib - good for teem.py, not good for C++
  • Weave - clear access to C++; can pass numpy arrays
  • Invoke CLI - clear distinction of C++ and python; easily supports multiple calling languages; CLIs are reusable executables/libraries; clear how to deploy; ITK IO requires one memcopy; approach is pretty indirect.
• Confirmation that weave is a promising approach, but CLI still seems to be preferred approach.

References

• http://www.slideshare.net/gsroma/integrando-c-com-python-presentation (in portugese)
• http://www.scipy.org/Cookbook/Weave