SemanticPOSS: A Point Cloud Dataset with Large Quantity of Dynamic Instances

Abstract

3D semantic segmentation is one of the key tasks for autonomous driving system. Recently, deep learning models for 3D semantic segmentation task have been widely researched, but they usually require large amounts of training data. However, the present datasets for 3D semantic segmentation are lack of point-wise annotation, diversiform scenes and dynamic objects. In this paper, we propose the SemanticPOSS dataset, which contains 2988 various and complicated LiDAR scans with large quantity of dynamic instances. The data is collected in Peking University and uses the same data format as SemanticKITTI. In addition, we evaluate several typical 3D semantic segmentation models on our SemanticPOSS dataset. Experimental results show that SemanticPOSS can help to improve the prediction accuracy of dynamic objects as people, car in some degree. SemanticPOSS will be published at \url{www.poss.pku.edu.cn}.

submited to IEEE Intelligent Vehicles Symposium(2020)

References in corpus (3)

• ScanNet: Richly-annotated 3D Reconstructions of Indoor Scenes
• PointSeg: Real-Time Semantic Segmentation Based on 3D LiDAR Point Cloud
• SynthCity: A large scale synthetic point cloud

Cited by in corpus (6)

• Towards Semantic Segmentation of Urban-Scale 3D Point Clouds: A Dataset, Benchmarks and Challenges
• Sparse Single Sweep LiDAR Point Cloud Segmentation via Learning Contextual Shape Priors from Scene Completion
• LiDARNet: A Boundary-Aware Domain Adaptation Model for Point Cloud Semantic Segmentation
• Are We Hungry for 3D LiDAR Data for Semantic Segmentation? A Survey and Experimental Study
• Scene Context Based Semantic Segmentation for 3D LiDAR Data in Dynamic Scene
• Learning to Reconstruct and Segment 3D Objects