FLS: Scale-Invariant Fast Functional Registration

This repositoriy contains implementaion of the functional least-squares (FLS) algorithm from to the paper Scale-Invariant Fast Functional Registration.

FLS is a point cloud registration algorithm with support for unknown scale, partial overlap, and varying densities. More demonstrations can found in the project website.

This repository contains implementation of FLS in two forms: (1) a C++ library that can easily used for computer vision tasks; (2) a JAX implementation with potential application in learning-based task.

If you find FLS helpful and use it your projects, please cite the project as below:

@INPROCEEDINGS{SunM-ISRR-22, 
    AUTHOR    = {Muchen Sun AND Allison Pinosky AND Ian Abraham AND Todd Murphey}, 
    TITLE     = {{Scale-Invariant Fast Functional Registration}}, 
    BOOKTITLE = {Proceedings of International Symposium of Robotics Research}, 
    YEAR      = {2022}, 
    ADDRESS   = {Geneva, Switzerland}, 
    MONTH     = {September}
}

C++ Implementation

The source code of the C++ implementation is under the directory fls-plusplus. Information regarding installation and usage can be found under the same directory.

JAX Implementation

The source code of the JAX implementation is under the directory fls-jax. Information regarding installation and usage can be found under the same directory. You can try the algorithm without installation via Colab.

Copyright and License

The implementations contained herein are copyright (C) 2021 - 2022 by Muchen Sun, and are distributed under the terms of the GNU General Public License (GPL) version 3 (or later). Please see the LICENSE for more information.

Contact: muchen@u.northwestern.edu

Lab Info: Prof. Todd D. Murphey, Interactive & Emergent Autonomy Lab / Center for Robotics and Biosystems / Northwestern University