README.md

Pose Regularization

This directory contains a Cython wrapper for the pose regularization library.

Building the Python extension

Numpy and Cython must be installed in order to build this extension. It is recommended to use a virtual environment, as usual.

virtualenv -p <PATH_TO_YOUR_INTERPRETER> <DESIRED_VIRTUALENV_LOCATION>
source <DESIRED_VIRTUALENV_LOCATION>/bin/activate

Once in the python directory, you can build and install the extension into your environment:

cd <REPO_ROOT>/python
pip install numpy cython
python setup.py bdist_wheel
pip install dist/pose_regularization-*.whl

The resulting wheel file in python/dist can be saved for later use or redistributed (to compatible systems).

Using the package

The algorithm is implemented as a function regularize_matrices_4x4, which takes as input a stack of 4x4 Euclidean transformation matrices and a set of parameters. The returned output is the regularization result, in the same format as the input.

The following example shows the application of the algorithm to dummy data. For the action of each parameter, please refer to the docstrings.

import numpy as np
from pose_regularization import regularize_matrices_4x4, Regularization as R

poses_noisy = np.array([np.eye(4)]*5)  # replace with your input signal

poses_regularized = regularize_matrices_4x4(poses_noisy, 
    R.L1, R.L2, R.HUBER, 0.25,
    50, 1, 100, 50) 

Testing the Python extension

If you have cython and numpy in your virtualenv, you can execute the tests from the python directory:

cd <REPO_ROOT>/python
python setup.py test