PointPillar based CenterPoint

Basis Codes: OpenPCDet and CenterPoint

• OpenPCDet is a clear, simple, self-contained open source project for LiDAR-based 3D object detection.
  • OpenPCDet repository is dedicated solely to inferencing the CenterPoint-pointpillar model, Center-based 3D Object Detection and Tracking.
  • I also improved the OpenPCDet with Mr. JongRok-Lee in the following repository:
    • https://github.com/JongRok-Lee/CenterPoint.git
    • https://github.com/hyunkoome/CenterPoint.git

Docker Environment

• Base Image: nvcr.io/nvidia/tensorrt:23.04-py3
• OS: Ubuntu 20.24
• CUDA: 12.1.0
• cuDNN: 8.9.0
• TensorRT: 8.6.1
• python: 3.8
• Pytorch: 2.1.1

1. Setting Dev Environment using Docker Env

• Please follow docs/1_Setting_Environment_using_Docker.md and proceed with the instructions.

2 Preparing Datasets

• Please follow docs/2_Preparing_Waymo_Dataset.md and proceed with the instructions.

3. Setting OpenPCDet.

• Please follow docs/3_Setting_OpenPCDet.md and proceed with the instructions.

4. Training a model.

• Please follow docs/4_Training.md and proceed with the instructions.

5. Testing a model and Evaluation with Pytorch models.

• Please follow docs/5_Testing_Evaluation.md and proceed with the instructions.

6. Convert ONNX models from Pytorch models

• Please follow docs/6_Convert_Models.md and proceed with the instructions.

7. Testing a model and Evaluation with TensorRT models.

• Please follow docs/7_Testing_Evaluation_with_TensorRT.md and proceed with the instructions.

8. Inference a model with TensorRT on ROS2 (Python)

• Please follow docs/8_Inference_ROS2_TensorRT_Python.md and proceed with the instructions.

9. Inference a model with TensorRT on ROS2 (C++)

• Please follow docs/9_Inference_ROS2_TensorRT_Cpp.md and proceed with the instructions.

------------------------------------------------------------------------------------------

TODO

• Fixed codes.
• Updated README.md files.
• Refactoring Phase1: Remove all codes not related to centerpoint-pillar.
  • Remove codes not related to network models (centerpoint-pillar).
  • Remove codes not related to waymo dataset.
  • Remove codes not related to functions.
  • Remove all parameters not related to centerpoint-pillar in the config.yaml files (networks and datasets).
  • Remove duplicated parameters in the config.yaml files (networks and datasets).
  • Merge duplicated codes of header, source (c++ and cuda) for pybind and ROS.
• Refactoring Phase2: Restructure the codes and config files.
  • Restructure all codes.
  • Re-arranged parameters and restructured on the config.yaml files of datasets and models.
  • Fix the code to create proper config.yaml when running export_onnx.py with cfg_file cfgs/waymo_models/centerpoint_pillar_inference.yaml.
• Refactoring Phase3: Add unit test functions for python, c++, cuda.