SimpleTrack: Understanding and Rethinking 3D Multi-object Tracking

3D multi-object tracking (MOT) has witnessed numerous novel benchmarks and approaches in recent years, especially those under the “tracking-by-detection” paradigm. Despite their progress usefulness, an in-depth analysis of strengths weaknesses is not yet available. In this paper, we summarize current MOT methods into a unified framework by decomposing them four constituent parts: pre-processing detection, association, motion model, life cycle management. We then ascribe failure cases existing algorithms to each component investigate detail. Based on analyses, propose corresponding improvements which lead strong simple baseline: SimpleTrack. Comprehensive experimental results Waymo Open Dataset nuScenes demonstrate that our final method could achieve new state-of-the-art with minor modifications. Furthermore, take additional steps rethink whether authentically reflect ability for real-world challenges. delve details find some intriguing facts. Finally, analyze distribution causes remaining failures SimpleTrack future directions MOT. Our code at https://github.com/tusen-ai/SimpleTrack .