UP3D: Utilizing Pseudo Points to Enhance 3D Object Detection

LiDAR-based 3D object detection is a critical engineering component in autonomous driving systems, yet its performance is fundamentally constrained by the sparsity and range-dependent density of point clouds, especially under sparse LiDAR observations caused by increasing sensing distance. Pseudo point assisted approaches alleviate this limitation by lifting image pixels into 3D space via depth completion; however, most existing methods exploit pseudo points at only a single stage of a two-stage detection pipeline, which restricts their ability to consistently complement sparse LiDAR observations. In this paper, we propose UP3D, a pseudo point assisted 3D object detection framework that integrates pseudo points into both the proposal generation and refinement stages in a stage-aware fashion. By leveraging pseudo points throughout the detection pipeline, UP3D densifies sparse LiDAR observations at the proposal stage to mitigate range-dependent point sparsity, and introduces complementary geometric cues at the RoI refinement stage for accurate localization, while maintaining computational efficiency through carefully designed pseudo point processing strategies. Experiments demonstrate that UP3D achieves competitive detection accuracy with favorable efficiency, validating the effectiveness of stage-aware pseudo point integration. Code: https://github.com/yujmo/UP3D .