TossherO/3D_Perception ? reverse-engineered prompt

Build me a working research repo for a LiDAR and camera fusion system that can detect pedestrians, cars, and cyclists in 3D, then track them and predict their future trajectories on the CODa dataset.

I want the project to be split into detection, tracking, and trajectory prediction so I can run each part separately, train or test the detection model, generate detection results for tracking, and train or test the trajectory prediction model. Please include simple commands or scripts for each step, use the CODa dataset layout described in the README, and make sure the demo folders can show example outputs and videos.

If anything is missing, set it up in a practical way and add clear instructions so I can download the dataset, put the checkpoints in the right place, and run the full pipeline end to end. Look up current docs online if you need to.