BEVFusion training procedure

[AinaraC]

I have a question about BEVFusion training. Why is it necessary to train the LiDAR-only model first and then use it for the fusion model training? MVX-Net can be trained directly for fusion