Abstract:With the rapid advances of autonomous driving, it becomes critical to equip its sensing system with more holistic 3D perception. However, existing works focus on parsing either the objects (e.g. cars and pedestrians) or scenes (e.g. trees and buildings) from the LiDAR sensor. In this work, we address the task of LiDAR-based panoptic segmentation, which aims to parse both objects and scenes in a unified manner. As one of the first endeavors towards this new challenging task, we propose the Dynamic Shifting Network (DS-Net), which serves as an effective panoptic segmentation framework in the point cloud realm. In particular, DS-Net has three appealing properties: 1) strong backbone design. DS-Net adopts the cylinder convolution that is specifically designed for LiDAR point clouds. The extracted features are shared by the semantic branch and the instance branch which operates in a bottom-up clustering style. 2) Dynamic Shifting for complex point distributions. We observe that commonly-used clustering algorithms like BFS or DBSCAN are incapable of handling complex autonomous driving scenes with non-uniform point cloud distributions and varying instance sizes. Thus, we present an efficient learnable clustering module, dynamic shifting, which adapts kernel functions on-the-fly for different instances. 3) Consensus-driven Fusion. Finally, consensus-driven fusion is used to deal with the disagreement between semantic and instance predictions. To comprehensively evaluate the performance of LiDAR-based panoptic segmentation, we construct and curate benchmarks from two large-scale autonomous driving LiDAR datasets, SemanticKITTI and nuScenes. Extensive experiments demonstrate that our proposed DS-Net achieves superior accuracies over current state-of-the-art methods. Notably, we achieve 1st place on the public leaderboard of SemanticKITTI, outperforming 2nd place by 2.6% in terms of the PQ metric.

DGPolarNet: Dynamic Graph Convolution Network for LiDAR Point Cloud Semantic Segmentation on Polar BEV

Panoptic-PolarNet: Proposal-free LiDAR Point Cloud Panoptic Segmentation

PCB-RandNet: Rethinking Random Sampling for LIDAR Semantic Segmentation in Autonomous Driving Scene

LiDAR-based Panoptic Segmentation via Dynamic Shifting Network

PolarNet: Accelerated Deep Open Space Segmentation Using Automotive Radar in Polar Domain

LiDAR Point Clouds Semantic Segmentation in Autonomous Driving Based on Asymmetrical Convolution

MotionBEV: Attention-Aware Online LiDAR Moving Object Segmentation with Bird's Eye View based Appearance and Motion Features

Hybrid CNN-LSTM Architecture for LiDAR Point Clouds Semantic Segmentation

LiDAR-based 4D Panoptic Segmentation via Dynamic Shifting Network

DPS-Net: Deep Polarimetric Stereo Depth Estimation

Panoptic-PHNet: Towards Real-Time and High-Precision LiDAR Panoptic Segmentation via Clustering Pseudo Heatmap

Knowledge Distillation from 3D to Bird's-Eye-View for LiDAR Semantic Segmentation

PSFE-Net: Semantic Segmentation Network for Airborne LiDAR Transmission Corridor Scenes Inspection

PolarBEVDet: Exploring Polar Representation for Multi-View 3D Object Detection in Bird's-Eye-View

LiDAR Panoptic Segmentation for Autonomous Driving

Location-Guided LiDAR-Based Panoptic Segmentation for Autonomous Driving.

SegNet4D: Efficient Instance-Aware 4D Semantic Segmentation for LiDAR Point Cloud

PARTNER: Level up the Polar Representation for LiDAR 3D Object Detection

Fully densely linked and strongly correlated instance segmentation for street-level images

EIDU-Net: edge-preserved inception DenseGCN U-Net for LiDAR point cloud segmentation

SFPNet: Sparse Focal Point Network for Semantic Segmentation on General LiDAR Point Clouds