Supervoxel for 3D point clouds

A python version of the supervoxel segmentation method for 3D point clouds. The original C++ code can be found here and here.

Three supervoxel segmentation method are implemented:

• VCCS: The original VCCS method
• VCCS with kNN: A variation of VCCS that uses kNN to find neighbors.
• yblin's method : ISPRS 2018 paper

Note that there has some problems in the implementation of VCCS with kNN (I will fix as soon as possible).

Install & complie

This repository requires C++11 to compile and python3 with pybind11 installed. Pybin11 can be installed by running pip install pybind11. Then, for python install, clone this repository, change to the root directory and run pip install ..

Sample usage:

Please see the demo.ipynb file for sample usage.

from vccs_supervoxel import segment,segment_knn,segment_vccs
#segment is the main function to call the supervoxel segmentation provided by yblin
#segment_knn is the main function to call the supervoxel segmentation provided by VCCS with kNN
#segment_vccs is the main function to call the supervoxel segmentation provided by VCCS
# Input should have size [N,9] i.e. [N, xyz rgb nxnynz] 
out = segment_vccs(pts[:, :9], 1,1) #or segment_knn(pts[:, :9], 1) or segment(pts[:, :9], 1) 
# out is 2-D NumPy array have size [N,10] containing xyz rgb labelcolor(rgb) label