A novel framework, MOSE, that leverages scene-specific features called "scene cues" to boost monocular 3D object detection performance on roadside cameras, outperforming state-of-the-art methods.
