Towards Learning Group-Equivariant Features for Domain Adaptive 3D Detection Sangyun Shin Yuhang He