Skip to content
A structured optimization framework for spatially regularizing point clouds classification
C++ MATLAB
Branch: master
Clone or download
loicland fixed the edge_weight reshaping
Thanks for @hehe549124  for pointing it out.
Latest commit 4bff4bf Jul 22, 2019
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
PFDR_simplex Add files via upload Jun 21, 2017
L0_cut_pursuit.m
LBP_max_product.m Add files via upload May 5, 2017
LBP_sum_product.m Add files via upload May 5, 2017
LICENSE.md Update LICENSE.md May 4, 2017
PFDR.m Add files via upload May 5, 2017
README.md
alpha_expansion.m Add files via upload May 5, 2017
benchmark.m made running the code clearer Aug 7, 2018
build_graph_structure.m fixed the edge_weight reshaping Jul 22, 2019
configure.m
data.tar.gz Add files via upload May 5, 2017
evaluate.m Add files via upload May 5, 2017
evaluate_partial.m Add files via upload May 5, 2017

README.md

point-cloud-regularization

A structured optimization framework for spatial regularization and segmentation of point clouds, with Matlab interface Loic Landrieu 2017

Regularization:

regularization

Based on:

A structured regularization framework for spatially smoothing semantic labelings of 3D point clouds. Landrieu, L., Raguet, H., Vallet, B., Mallet, C., & Weinmann, M. (2017).

This framework propose a set of methods for spatialy regularizing semantic labelings on a point cloud. As mentioned in the paper above, 4 fidelity functions and 3 regularizers are proposed.

Segmentation:

segmentation

Based on:

Weakly supervised segmentation-aided classification of urban scenes from 3D LiDAR point clouds. Guinard, S., & Landrieu, L. In ISPRS 2017

Fast segmentation of point clouds with L0-cut pursuit.

DEPENDENCIES:

CUT PURSUIT : https://github.com/loicland/cut-pursuit

PFDR : From https://github.com/1a7r0ch3

ALPHA-EXPANSION : http://vision.ucla.edu/~brian/gcmex.html

LOOPY BELIEF PROPAGATION : http://www.cs.ubc.ca/~schmidtm/Software/UGM.html

All those dependencies are optional, but access to the corresponding regularization are dependant on which ones are installed. If you chose not to install some of those libraries, some code commenting might be necessary.

The data compressed files needs to be dezipped. All credits goes to http://www.cs.cmu.edu/~vmr/datasets/oakland_3d/cvpr09/ for the data.

RUNNING THE CODE:

Run the lines from configure.m corresponding to the method you are interested to try.

Follow benchmark.m to see examples of calls.

You can’t perform that action at this time.