Recent works recognized lidars as an inherently streaming data source and showed that the end-to-end latency of lidar perception models can be reduced significantly by operating on wedge-shaped point cloud sectors rather then the full point cloud. However, due to use of cartesian coordinate systems these methods represent the sectors as rectangular regions, wasting memory and compute. In this work we propose using a polar coordinate system and make two key improvements on this design. First, we increase the spatial context by using multi-scale padding from neighboring sectors: preceding sector from the current scan and/or the following sector from the past scan. Second, we improve the core polar convolutional architecture by introducing feature undistortion and range stratified convolutions. Experimental results on the nuScenes dataset show significant improvements over other streaming based methods. We also achieve comparable results to existing non-streaming methods but with lower latencies. The code and pretrained models are available at \url{https://github.com/motional/polarstream}.
LiDAR-based Online 3D Video Object Detection with Graph-based Message Passing and Spatiotemporal Transformer Attention
Beijing Institute of Technology（北京理工大学）；Baidu Research（百度研究院）；Inception Institute of Artificial Intelligence（人工智能起源研究所）；National Engineering Laboratory of Deep Learning Technology and Application（深度学习技术及应用国家工程实验室）；University of Kentucky（肯塔基大学）
Improved Pillar with Fine-grained Feature for 3D Object Detection
Beihang University（北京航空航天大学）；YiGo Robotics（易控机器人）
基于LiDAR点云的三维目标检测在自动驾驶感知模块中具有重要的作用，对检测速度、稳定性和准确性有较高的要求。然而，现有的基于点的方法由于原始点太多而难以达到速度要求，而基于体素的方法由于3D稀疏卷积而无法保证稳定的速度。相比之下，基于二维网格的方法，如PointPillar，可以很容易地实现一个稳定的和高效的速度基于简单的二维卷积，但它很难得到竞争力的精度受到粗粒度的点云表示。因此，本文在PointPillar的基础上提出了一种具有细粒度特征的改进型柱子，可以显著提高柱子的检测精度。该模型由高度子柱和稀疏微柱两个模块组成，分别在三维空间的垂直和水平方向上实现了细粒度的表示。对于高度敏感的子柱，我们引入了高度位置编码，以保持每个子柱的高度信息，在投影到一个2D伪图像。对于稀疏型微柱模型，我们引入了基于稀疏的CNN骨干，由密集特征和稀疏注意力模块堆叠而成，以有效地提取具有更大感受野的特征。实验结果表明，我们提出的方法在Waymo Open Dataset上的性能明显优于以前最先进的3D检测方法。相关规范将发布，以方便学术界和工业界的研究。
Enhancing Lidar-based Object Detection in Adverse Weather using Offset Sequences in Time
Institute for Automotive Engineering RWTH Aachen University（亚琛工业大学汽车工程研究所）；Institute of Imaging and Computer Vision RWTH Aachen University（亚琛工业大学图像与计算机视觉研究所）
Multistream Network for LiDAR and Camera-based 3D Object Detection in Outdoor Scenes
Department of Computer Science, The University of Western Australia（西澳大学计算机科学系）；School of Computing & Information Systems, The University of Melbourne（墨尔本大学计算与信息系）
Real-Aug: Realistic Scene Synthesis for LiDAR Augmentation in 3D Object Detection
Inspur Electronic Information Industry Co.,Ltd.（浪潮电子信息产业股份有限公司）；Centre for Artificial Intelligence and Robotics, HKISI, CAS（中国科学院香港创新研究院人工智能与机器人中心）；Institute of Automation, CAS（中国科学院自动化研究所）
PAN: Pillars-Attention-Based Network for 3D Object Detection
Department of Electronic and Computer Engineering, Lero (the Research Ireland Centre for Software), and the Data Driven Computer Engineering (D 2 iCE) Research Centre at the University of Limerick（电子与计算机工程系，Lero（爱尔兰软件研究中心），以及利默里克大学数据驱动计算机工程（D 2 iCE）研究中心）；Provizio, Future Mobility Campus Ireland, Shannon Free Zone（Provizio，Future Mobility Campus Ireland，Shannon Free Zone）
ContrastMotion: Self-supervised Scene Motion Learning for Large-Scale LiDAR Point Clouds
Nanjing University of Aeronautics and Astronautics（南京航空航天大学）；Rhino（犀牛（公司））；The Chinese University of Hong Kong（香港中文大学）；OPPO Research Institute（OPPO研究院）；Zhejiang Lab（之江实验室）；ShanghaiTech University（上海科技大学）
PillarTrack:Boosting Pillar Representation for Transformer-based 3D Single Object Tracking on Point Clouds
School of Mathematics and Computer Science, Nanchang University（南昌大学数学与计算机科学学院）；School of Automation, Southeast University（东南大学自动化学院）；Aberdeen Institute of Data Science and Artificial Intelligence, South China Normal University（华南师范大学 Aberdeen 数据科学与人工智能研究院）；Houmo AI
Comprehensive Robustness Analysis of LiDAR-based 3D Object Detection in Autonomous Driving
Institute for TMDT, University of Wuppertal, Germany（TMDT研究所，乌尔姆大学，德国）；IKB Faculty of Science, University of British Columbia, Canada（科学学院，不列颠哥伦比亚大学，加拿大）
Multi-Correlation Siamese Transformer Network with Dense Connection for 3D Single Object Tracking
Zhengzhou University（郑州大学）；Institute of Automation, Chinese Academy of Sciences（中国科学院自动化研究所）；University of Chinese Academy of Sciences（中国科学院大学）；Nullmax Inc（Nullmax公司）
