Abstract

This paper introduces MCTrack, a new 3D multi-object tracking method thatachieves state-of-the-art (SOTA) performance across KITTI, nuScenes, and Waymodatasets. Addressing the gap in existing tracking paradigms, which oftenperform well on specific datasets but lack generalizability, MCTrack offers aunified solution. Additionally, we have standardized the format of perceptualresults across various datasets, termed BaseVersion, facilitating researchersin the field of multi-object tracking (MOT) to concentrate on the corealgorithmic development without the undue burden of data preprocessing.Finally, recognizing the limitations of current evaluation metrics, we proposea novel set that assesses motion information output, such as velocity andacceleration, crucial for downstream tasks. The source codes of the proposedmethod are available at this link:https://github.com/megvii-research/MCTrack}{https://github.com/megvii-research/MCTrack