An approach to 3D Object Tracking which is evaluated on segments on the Waymo Open Dataset. The approach is heavily influenced by Weng's Baseline for 3D Multi-Object Tracking.
Create a virtual environment and install dependencies
python3 -m venv venv
source venv/bin/active
pip install -r requirements.txt
Download a dataset segment (.tfrecord file) and run the tracker evaluation with
python main.py \
--tracker predictive \
--segment path/to/segment.tfrecord \
--dropout 0.2