Has MapQR been tested with multi-modal inputs (LiDAR/Image)

[Han-MH]

Hi, thanks for your great work on MapQR!

I’m trying to extend MapQR to a multi-modal setting (e.g., radar + image).
However, I encountered an issue: when adding the LiDAR branch, the training loss becomes even lower than the original MapQR, but the final mAP drops significantly.

Before diving further, I’d like to ask:
Have you ever experimented with a multi-modal version of MapQR (e.g., LiDAR+image or radar+image)?

If so, do you have any suggestions or best practices for fusing additional BEV features into the MapQR framework?

Any insights or recommendations would be greatly appreciated.
Thanks again for releasing this excellent work!