trautmane/EM_aligner_python

Python version of Khaled Khairy's EM_aligner, supporting distributed assembly and solve

EM_aligner_python

Alignment of EM datasets. Builds sparse matrices from tilespecs and pointmatch collections and solves for new transforms using constrained linear least squares.

Level of support

We are planning on occasional updating this tool with no fixed schedule. Community involvement is encouraged through both issues and pull requests.

setup

1. clone and cd into the repo directory
2. modify sourceme.sh to point RENDER_JAVA_HOME to a directory where bin/java can be found
3. source sourceme.sh

alternately, ignore sourceme.sh and change the client_scripts in the input jsons to point to your own version of the render scripts

montage

align the images of each tissue section to each other. sections are treated independently.

for z in zlist:
    assemble from montage_matches(z)
    solve with scipy.sparse
    ingest

command line from within EMaligner directory:

::

python -m EMaligner.EMaligner --input_json /path/to/montage_test.json

rough alignment

apply a 3D alignment to downsampled montage results. Each montaged section is treated as 1 tile. The tiles are aligned with a rigid transformation. The first tile is fixed to prevent drastic scaling.

command line:

python -m EMaligner.EMaligner --input_json example_jsons/rough_test.json

3D alignment

all tiles are treated separately,with affine transformations. ## small 3D run 3D solve locally, will be RAM limited

for z in zlist:
    add to assembly from montage_matches(z), cross_matches(z,z+1),...,cross_matches(z,z+depth)
solve with scipy.sparse
ingest

command line:

python -m EMaligner.EMaligner --input_json example_jsons/small3D_test.json

large 3D

intended to assemble A locally into multiple files for input to a distributed solver

for z in zlist:
    add to assembly from montage_matches(z), cross_matches(z,z+1),...,cross_matches(z,z+depth)
    if assembly>big:
        write_new_file
        assembly = 0
solve elsewhere with PETSc
ingest

command line:

python not_ready_yet.py --input_json not_ready_yet.json
tar cvzf the_files
scp the_files.tar.gz user@big_machine:
user@big_machine$ tar -xvzf the_files.tar.gz
user@big_machine$ solve the_files
scp user@big_machine:the_solution ./
python ingest.py the_solution