PillarTsAE: A High-Performance Pillar-based 3D Object Detection Network in LIDAR Point Clouds

Engineering notes

Sufficient experiments demonstrate that our method outperforms other models, achieving state-of-the-art performance on KITTI with mAP, showcasing its powerful and robust ability.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

3D object detection plays a crucial role in many fields such as autonomous driving, robot perception and other fields. Current methods encounter limitations when dealing with intricate point cloud data, such as poor performance in detecting small objects and the high computational overhead they exhibit. This study proposes a high-performance network dubbed PillarTsAE, addressing problems above effectively. Specifically, it consists of two novel modules: TsAFE and PPFE, enhancing the expression of point cloud features and examining both inter-pillar and intra-pillar relational features, respectively. Finally, PillarTsAE is equipped with BiFPN in neck module. Sufficient experiments demonstrate that our method outperforms other models, achieving state-of-the-art performance on KITTI with mAP, showcasing its powerful and robust ability.

Links and sources

• Official / arXiv page

Comments