A 3D object tracker for pre-computed detections, currently based on Kalman Filter.
We strongly recommend to use conda environments since the dataset APIs require incompatible python versions!
If you don't have any pre-computed detection results run an example for on-the-fly generated fake data.
python run.py fake
For the Argoverse dataset you may download the detection results of coming soon.
Use wildcards to define a list of input files.
Replace $DATAROOT
with where ever you used to store all the data.
python run.py argoverse \
--inputfiles '$DATAROOT/results/agroverse_detections_2020/validation/*/*/*.json' \
--groundtruth '$DATAROOT/argoverse-tracking/val/*/per_sweep_annotations_amodal/*.json'