seems a little heavy.

MSCE is implemented as a single-layer Transformer encoder

起了个好名字

ObjectFusion: Multi-modal 3D Object Detection with Object-Centric Fusion, iccv23