Abstract: Accurate and reliable 3D object detection is crucial for autonomous driving, normally achieved using camera-only or camera-LiDAR fusion methods based on BEV (Bird’s Eye View) perspective.