GitLab

* support train on nus

* refactor transfusion head

* img branch optioinal

* support nuscenes_mini in replace_ceph_backend

* use replace_ceph

* add only-lidar

* use valid_flag in dataset filter

* support lidar-only training 69

* fix RTS

* fix rotation in ImgAug3D

* revert to original rotation in ImgAug3D

* add LSSDepthTransform and parse_losses

* fix LoadMultiSweeps

* fix bug about points in-place operations

* support amp and replace syncBN by BN

* add amp config

* set growth-interval in amp

* Revert "fix LoadMultiSweeps"

This reverts commit ab27ea1941f80c5d7fb3071005109f4794c9c9cb.

* add float in cls loss

* iter_based lr in fusion stage

* rename config

* use normalization query pos for stable training

* remove unnecessary code & simplify config & train 5 epoch

* smaller ete_min_ratio

* polish code

* fix UT

* Revert "use normalization query pos for stable training"

This reverts commit 30091188d4a665beaadcd5a9b49e04c9da47139e.

* update readme

* fix height offset

* init centerformer in projects

* add readme and disable_tf32_switch

* using our iou3d and nms3d, using basicblock in mmdet, simplify code in projects

* remove attention.py and sparse_block.py

* only using single fold

* polish code

* polish code and add dosstring

* add ut for disable_object_sample_hook

* modify data_root

* add ut

* update readme

* polish code

* fix docstring

* resolve comments

* modify project names

* modify project names and add _forward

* fix docstring

* remove disable_tf32