AB3DMOT

Why we included this project

AB3DMOT implements the baseline from a widely cited IROS 2020 paper on 3D multi-object tracking, so it is a natural first stop for anyone working on autonomous driving or robotics perception. The method runs a Kalman filter over per-frame 3D detections and links them into consistent trajectories, which keeps the whole pipeline light enough to adapt or benchmark against before committing to a heavier learned tracker. The repo ships KITTI evaluation tools, later added nuScenes support, and includes the metrics the paper introduced, so reproducing the published results and measuring your own runs is straightforward. Because the code is deliberately small and modular, it also works well as readable reference material for how a classic tracking pipeline is put together. One caveat: the code is released under Carnegie Mellon's noncommercial academic license.