Institute of Information Science, Academia Sinica

The realm of autonomous vehicles places a spotlight on the importance of precise 3D object detection, with LiDAR technology standing as a critical contributor. This talk discusses the innovative advancements in LiDAR point cloud upsampling and 3D object detection. The first work, "Density-imbalance-eased LiDAR Point Cloud Upsampling via Feature Consistency Learning," presents a robust methodology to counter the challenges of density variation in LiDAR scans, ensuring the consistency of features during the upsampling process. Building on this foundation, our subsequent work, "Singular Value Control and Distance Alignment Network for 3D Object Detection," introduces an approach that blends the critical attributes of singular value control with distance alignment, enhancing accuracy in 3D object detection tasks. Together, these publications illuminate the pathway toward a better LiDAR-based solution for LiDAR point cloud processing and 3D object detection.