GitLab

* Support to train using FP16

* fix type inconsistency error on naive syncBN

* resolve comments

* clean nan check

* Add centerpoint_rpn and scn, change pillar encoder and voxel_encoder

* Move test_voxel_encoders.

* Change names, add docstring.

* Reconstruct centerpoint_rpn.

* Add centerpoint_rpn.

* Change SECONDFPN, delete centerpoint_fpn

* Remove SparseBasicBlock.

* Change SpMiddleResNetFHD to SparseEncoder.

* Finish SparseEncoder unittest.

* Change test_hard_simple_VFE.

* Change option, add legacy.

* Change docstring, change legacy.

* Fix legacy bug.

* Change unittest, change docstring.

* Change docstring.