mistycheney / mousebrainatlas Goto Github PK

-Unable to load or save contours (KeyError: 'orientation')
-Unable to interpolate contours (KeyError: 'position_um')
No special conditions needed to reproduce either of these

Allow users to specify which are the bad sections, and take proper actions.

• For example, if section 3 is a bad section and can be discarded, then we can choose to replace it with section 2 or section 4. This means skipping aligning section 3 to section 2, and register section 4 to section 2 instead of to section 3.
• If we cannot discard section 3 but it causes elastix failure, then manually align section 3 to section 2 and/or align section 4 to section 3. Feed the manually obtained transform parameters to algorithm.

I was thinking about the issue of how to easily distribute the software.

I think a good way would be to follow the standard convention where, after your clone the github directory, you go to the root directory and run "python setup.py install" which defines the different data directories and also sets environment variables that point to where the data is on gordon etc.

Then there should be scripts for uploading to gordon, setting the parameters for the gabor filters etc, downloading results and then opening these results in the GUI.

I tried to refactor the code, so that there are FeatureExtractor class and ModelDetector class. However calling joblib Parallel functions within class methods causes "can't pickle instancemethod".

possible solution: use new serialization method, such as dill, which is implemented within pathos
reference: http://stackoverflow.com/questions/19984152/what-can-multiprocessing-and-dill-do-together

joblib is not yet using dill, although it is on their todo list (https://github.com/joblib/joblib/search?q=dill&ref=cmdform&type=Issues). So I tried to use pathos.multiprocessing directly,
from pathos.multiprocessing import ProcessingPool as Pool

Installation of pathos: official website is http://trac.mystic.cacr.caltech.edu/project/pathos, but that release is very stale. Instead, use the development version https://github.com/uqfoundation/pathos and follow the instructions in the repo

This manages to do feature computation perfectly, but degrades to using only 2 or 3 processes at texton computation. I have no idea why.

checkin: test_feature_extraction_pathos_failed.py

• Allow cross-reference in virtual coronal / horizontal sections
• Speed up image loading by parallelizing. (Increase workstation RAM to the maximum 196GB = 210 sections in RAM ~ 90% brainstem-to-brainstem sections in one brain.)
• Allow the user to choose between showing 1/2/3 panels.
• Show section number on the left-upper corners of images.