Event-LiDAR: 3D Eventification for Efficient Point Cloud Processing
ECCV 2026 · Efficient 3D object detection with temporally sparse LiDAR point clouds.
Event-LiDAR converts consecutive LiDAR scans into temporally sparse point clouds by retaining geometric changes that a first-order model cannot explain. On nuScenes (1F+9T), it reduces the number of points by 72% while maintaining accuracy comparable to full-scan baselines, with 23% faster end-to-end inference on PTv3.