AutoExpert: Automating 3D LiDAR Annotation from Expert-Crafted Guidelines
AutoExpert is a new benchmark that tests multimodal few-shot learning for 3D detection in LiDAR data without needing manual 3D annotations. It uses foundation models to label LiDAR data based on expert guidelines, mimicking human annotation. The method successfully generates accurate 3D cuboids by combining 2D object detection with VLM-guided multi-hypothesis testing.