This document provide tutorials to develop CenterPoint.
TODO
-
If you interested in developing tracking algorithms based on our detection results, please refer to NUSC and WAYMO. For more advanced tasks like motion prediction, you may need to store the final bev feature map computed here.
-
You will also need to add files to
det3d/datasets/pipelines/preprocess.py
to specify the data generation during training and training. -
You may also need to change the collate function in collate.py and data_loading function in trainer.py
Please add any 3D backbone in det3d/models/backbones
, any 2D backbones in det3d/models/necks
, and any two-stage refinement modules in det3d/models/second_stage
.
If you have any suggestions for improving this codebase for development, please open an issue or send us an email.