SemanticKITTI is designed for research on laser-based semantic segmentation. The authors annotated all sequences of the KITTI Vision Odometry Benchmark and provide dense point-wise annotations for the complete 360 degree field-of-view of the employed automotive LiDAR. The original KITTI dataset showed inner city traffic, residential areas, but also highway scenes and countryside roads around Karlsruhe, Germany. The original odometry dataset consists of 22 sequences, splitting sequences 00 to 10 as training set, and 11 to 21 as test set. For consistency with the original benchmark, the authors adopt the same division for our training and test set. Moreover, they do not interfere with the original odometry benchmark by providing labels only for the training data. Overall, they provide 23,201 full 3D scans for training and 20,351 for testing, which makes it by a wide margin the largest dataset publicly available.
The authors labeled each scan resulting in a sequence of labeled point clouds, which were recorded at a rate of 10 Hz. This enables the usage of temporal information for semantic scene understanding and aggregation of information over multiple scans. They also annotated moving and non-moving traffic participants with distinct classes, including cars, trucks, motorcycles, pedestrians, and bicyclists. This enables to reason about dynamic objects in the scene.
@inproceedings{behley2019iccv,
author = {J. Behley and M. Garbade and A. Milioto and J. Quenzel and S. Behnke and C. Stachniss and J. Gall},
title = {SemanticKITTI: A Dataset for Semantic Scene Understanding of LiDAR Sequences},
booktitle = {Proc. of the IEEE/CVF International Conf.~on Computer Vision (ICCV)},
year = {2019}
}
