Input/output shapes of the BEVFusion.voxelize() method

[maxvandenhoven]

Thanks for the amazing work on BEVFusion!

I was trying to understand the code for the LiDAR encoder to re-use it in my own project, but I am having some issues with getting the entire project to run in order to inspect the tensor shapes. Could you help me with the input and output shapes of the BEVFusion.voxelize() method? I am particularly interested in the shape of the points input argument and the feats, coords, and sizes outputs. A brief description of what the dimensions are supposed to represent would also be greatly appreciated!