SemanticKITTI

SemanticKITTI是一个基于KITTI Vision Benchmark的大型数据集，提供KITTI Odometry Benchmark所有序列的密集点级语义标签，用于激光雷达基础的全景分割训练和评估。该数据集包含28个类别，区分移动和非移动物体，通过半自动过程生成时间上一致的实例注释，适用于车辆、行人、自行车手和摩托车手等交通参与者。数据集的创建旨在推动激光雷达基础的全景分割研究，提供在线评估平台，支持研究人员开发和比较新的算法。

SemanticKITTI is a large-scale dataset built upon the KITTI Vision Benchmark. It provides dense point-level semantic labels for all sequences of the KITTI Odometry Benchmark, serving as a resource for training and evaluating LiDAR-based panoramic segmentation tasks. This dataset encompasses 28 categories, distinguishing between moving and non-moving objects. It generates temporally consistent instance annotations via a semi-automatic process, covering traffic participants such as vehicles, pedestrians, cyclists, and motorcyclists. The dataset was developed to advance LiDAR-based panoramic segmentation research, and it offers an online evaluation platform to support researchers in developing and comparing novel algorithms.

搜集汇总

背景与挑战

背景概述

SemanticKITTI是一个基于KITTI Vision Benchmark的大规模LiDAR序列数据集，用于语义场景理解研究。它提供了完整的Odometry Benchmark序列标注，覆盖360度视角的自动驾驶激光雷达扫描，并以10Hz频率记录为连续点云序列，支持时间信息利用。数据集还标注了动态交通参与者（如汽车、行人等），适用于动态物体分析和多种任务，如语义分割、全景分割和移动物体分割。