Generalized-Shuffled-Linear-Regression

Code for the ICCV 2021 paper: Generalized Shuffled Linear Regression.

Authors: Feiran Li, Kent Fujiwara, Fumio Okura, and Yasuyuki Matsushita

UPDATE: The objective in Eq.7 should be corrected to . The following analysis remain the same.

1. Run the demo

For image registration and point cloud registration:

• Install necessary dependencies: $ pip3 install requirements.txt
• Run img_registration.py for the image registration demo, and pcd_registration.py for the point cloud registration one.

For isometric shape matching:

• Direct to the shape_matching folder in Matlab. R2019a or later is needed to use the matchpairs function to solve the linear assignment problem.
• Run demo.m for fun.
• We have used the orientation-preserving operator proposed in the excellent work BCICP, and this code is based on its release. Please pay attention to citation.
• If you wish to use ur own data, I have implemented a python wrapper of the fast-marching algorithm to compute geodesics of meshes.

2. Related works

Other implementations of shuffled linear regression:

• Linear Regression with Shuffled Labels
• Stochastic Expectation-Maximization for Shuffled Linear Regression

Techniques for speed up:

The main limitation of our current implementation lies in time efficiency, which is dominated by the LAP solver. Some CUDA-based Hungarian algorithms like this and this may help to address this problem.

3. Contact

Please feel free to raise an issue or email to [email protected] if you have any question regarding the paper or any suggestions for further improvements.

4. Citation

If you find this code helpful, thanks for citing our work as

@inproceedings{li2021gslr,
title = {Generalized Shuffled Linear Regression},
author = {Feiran Li and Kent Fujiwara and Fumio Okura and Yasuyuki Matsushita},
booktitle = {IEEE/CVF International Conference on Computer Vision (ICCV)},
year = {2021}
}