Skip to content

Latest commit

 

History

History
43 lines (37 loc) · 2.75 KB

README.md

File metadata and controls

43 lines (37 loc) · 2.75 KB
Raw

L1-Skeleton

Sample skeleton

This repo implements a compatibility wrapper for the L1-Skeletonization method atop https://github.com/HongqiangWei/L1-Skeleton, allowing for .ply files to be processed into Cloud Volume skeletons. L1-Skeletonization is a robust and efficient point-cloud based method, and may be useful as an alternative to volumetric methods such as Kimimaro.

Usage guide

Prerequisites

  • Build the container using
docker build -t pointcloudl1 - < Dockerfile
  • The program can be run interactively/graphically by forwarding a X-Server socket into the container (run the PointCloudL1 binary with no flags); see #3 (comment)

CLI

  • Generate the .skel file for any isotropic .ply by running
./pointcloudl1.sh <ply input path> <skel output path> <path to config json>
  • l1_skeleton.py provides various convenience functions to convert the np arrays into .ply files, .skel files into Cloud Volume skeletons, a Python function to wrap the call to the shell script, etc.
  • generate_skeleton.py provides an example of how to run multiple skeletonization jobs concurrently
  • viewer.py demonstrates how to visualize the point clouds and skeletons

Hyperparameters

The two most important settings are Down Sample Num (which specifies the number of points to approximate the skeleton) and Init Radius Para (specifies the initial size of the expanding sphere). The GUI automatically calculates reasonable settings for Init Radius Para, but the CLI requires you to manually specify it.

In practice, I keep Init Radius Para constant and downscale the input point cloud by an appropriate factor.

Tips

  • Adding random noise to the input point cloud aids the convergence of the skeletonization on "barely-disconnected" components; see this
  • The repo can be compiled natively on Arch Linux; see #2

Citing

Please cite the original paper:

@article{huang2013l1,
  title={L1-medial skeleton of point cloud.},
  author={Huang, Hui and Wu, Shihao and Cohen-Or, Daniel and Gong, Minglun and Zhang, Hao and Li, Guiqing and Chen, Baoquan},
  journal={ACM Trans. Graph.},
  volume={32},
  number={4},
  pages={65--1},
  year={2013}
}