RANSAC fits models to noisy data.

It is most useful when the data contains outliers that should be ignored completely during model fitting (Wikipedia RANSAC page).

import ransac
my_ransac = ransac.Ransac(
    model, num_sample_points, min_inliers, inlier_threshold, stop_iterations)
results = my_ransac.run(list_of_data_points)
print(results.fit, results.inliers, results.outliers)

While ransac.Ransac always fits a single model, ransac.XRansac or ransac.JLinkage can be used to detect and fit multiple underlying models. The number of models does not need to be specified in advance. XRansac is faster but uses additional parameters.

See this IPython notebook for more complete examples.

To run the IPython notebooks locally:

make ipython

Note that the various RANSAC threshold parameters can have a large effect on the results and will likely need to be tuned for any particular use case. In particular, inlier_threshold determines what is considered an inlier, while stop_iterations presents a trade off between accuracy and speed. If the number of outliers can be estimated, ransac.calculate_ransac_iterations can help choose a good stop_iterations value.

To use a custom model, subclass ransac.Model and implement the fit and predict methods. See ransac/models/base.py for more details.

Requires Python 3 and Make.

To install:

make

To run tests:

make test

This project is licensed under the MIT License - see the LICENSE.txt file for details.

This paper is the basis of the XRansac variant (no affiliation):

Zhang, W., Kosecká, J.: Nonparametric estimation of multiple structures with outliers. In: Vidal, R., Heyden, A., Ma, Y. (eds.) WDV 2006. LNCS, vol. 4358, pp. 60–74. Springer, Heidelberg (2006) PDF

This paper is the basis of the J-linkage variant (no affiliation):

Toldo, R., & Fusiello, A. (2008, October). Robust multiple structures estimation with j-linkage. In European conference on computer vision (pp. 537-547). Springer, Berlin, Heidelberg. PDF