What is SynPEB?
The Synthetic Plane Extraction Benchmarking Dataset was created in order to provide a benchmark for plane segmentation algorithms based on 3-D lidar data. The SynPEB world consists of a room of approximately 6 x 7 x 3 m populated with various polyhedral objects, resulting in 42.6 planes of different shapes and sizes per scan. Analogously to SegComp, the dataset is divided into 10 training scans and 30 testing scans, provided as organized point clouds of 500 x 500 measurements.
The image below shows a rendered view of the virtual room where all SynPEB scans were captured.
Where to download SynPEB?
How is SynPEB structured?
The SynPEB dataset contains a test and a training folder. Both folders contain five subfolders: gt, var1, var2, var3, and var4. The folder gt contains the ground-truth lidar scans, while the other folders contain noisy scans. The radial noise is normally distributed with a constant standard deviation of 20 mm, while the standard deviation of the angular noise amounts to 0.5 mdeg, 1 mdeg, 2 mdeg, and 4 mdeg for var1 to var4, respectively. The point clouds comprise 500 x 500 measurements each. They come in the form of compressed binary PCD files. Each point has four coordinates: x, y, z, and intensity. The intensity coordinate encodes the ground-truth ID of the plane that reflects the laser ray.
How to cite SynPEB?
If you use SynPEB, please cite our paper:
@INPROCEEDINGS{schaefer19icra,
author={Alexander Schaefer and Johan Vertens and Daniel B{\"u}scher and Wolfram Burgard},
booktitle={2019 IEEE/RSJ International Conference on Robotics and Automation (ICRA)},
title={A Maximum Likelihood Approach to Extract Finite Planes from 3-{D} Laser Scans},
year={2019},
month={May},
organization = {IEEE},
url={http://ais.informatik.uni-freiburg.de/publications/papers/schaefer19icra.pdf}
}
License
SynPEB is published under GNU General Public License, Version 3. In order to view the license file, please download the dataset.