Rethinking Dimensionality Reduction in Grid-based 3D Object Detection