• ROS2(humble, foxy)
• vision-msgs
• vision-msgs-rviz-plugins
• tf-transformation
• sensor-msgs
• mmdet3d(temporary) 1.0.0-rc6

rosdep install --from-paths src --ignore-src -r -y

Note:

If you are using Foxy distro instead of Humble, you need to manually build these dependencies:

• vision-msgs
• vision-msgs-rviz-plugins

In the patch 0.0.2, we introduced the ros2 params feature in order to avoid verbose build. You can rewrite the config file config.yaml in conf/ directory.

1. pub_nuscenes_bin

   • time_period

     Time period for releasing the point cloud. In nuscenes source, it is the frequency of publishing point cloud; in real-time lidar scan, it is the scan frequency.

   • nus_lidar_path

     The path to the pointcloud folder. It should end with 'LIDAR_TOP'.

2. infer_node

   • score_threshold

     The threshold of score for filtering the detection result.

   • mmdet3d_path

     The absolute path to the mmdet3d package.

   • config_file

     The relative path of config file in mmdet3d path.

   • checkpoint_file

     The absolute path to the checkpoint file.(.pth)

   • onnx_model(supported in v1.0.x)

     The absolute path to the onnxruntime model file.(.onnx)

colcon build --packages-select pc_det
. install/setup.bash

1. base demo (mmdet3d api)

ros2 launch pc_det nuscenes_demo_launch.py

2. ort-gpu infer (still bug)

ros2 launch pc_det nuscenes_ort_launch.py

3. direct pc2 format inference(skip the file transform)

ros2 launch pc_det nuscenes_pc_launch.py

4. run depthcam detection demo

ros2 launch pc_det depthcam_demo_launch.py