About output of backbone

[Dikui]

Hello, may I ask why only the fused lidar features are used in the case of transformer decoder join