ESAFusion: LiDAR--4-D Radar Fusion via Local Geometric Complementation and Multiscale Adaptive Interaction for 3-D Object Detection
The episode discusses ESAFusion, a framework for fusing LiDAR and 4-D radar data for 3-D object detection. The hosts detail its three-stage process: refining radar evidence, using local geometry for pillar complementation, and adapting feature weighting across scales. Results show high accuracy and robustness against sensor degradation in fog.