PrymceQ / nuScenes_TrajectoryVisualizationTool

The visualization project of the BEV model inference result, including detBox and trackingLine!

nuScenes_TrajectoryVisualizationTool

The visualization project of the BEV model inference result on NuScenes Dataset, including detBox and trackingLine!

First of all, I need to thank some Github authors! PF-Track | SimpleTrack | StreamPETR

We use Nuscenes v1.0-mini DATASET for example

The green dashed line represents the GT trajectory; solid lines represent predicted objects and trajectories, and the same color represents the same predicted object.

🌵Necessary File Format

• data/nuscenes/
  • v1.0-mini/
    • maps/
    • samples/
    • sweeps/
    • v1.0-mini/
• projects/
• tools/
• SimpleTrack/

🌵Build Envs

You can refer to the PF-Track configuration environment documentation.

Or use the Conda env configuration file we provide.

conda env create -f nuScenesTrajectoryVisualizationTool_env.yaml

as for mot_3d package,

cd SimpleTrack
pip install -e ./

🌵PKL Create

Like othor bev models, you need to use create_data.py to create pkl files for v1.0mini:

tracking_forecasting-mini_infos_train.pkl

tracking_forecasting-mini_infos_val.pkl

python tools/create_data.py nuscenes-tracking --root-path data/nuscenes/v1.0-mini --out-dir data/nuscenes/v1.0-mini --extra-tag tracking_forecasting --version v1.0-mini --forecasting

🌵Data Prepare

The final json data structure should like this, it should be notice that key tracking_id is necessary.

You can use the code here to prepare the json file with tracking_id key from the test result json file.

🌵Camera Visualization Code

tools/camera_visualization.py --result mini_track.json --show-dir work_dirs/cam_visualization/

🌵BEV Visualization Code

tools/bev_traj_visualization.py projects/configs/tracking/petr/f3_q500_800x320.py --result mini_track.json --show-dir work_dirs/tracking_visualization/