Reproduce “Immortal Tracker: Tracklet Never Dies” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Qitai Wang, Yuntao Chen, Ziqi Pang, Naiyan Wang, Zhaoxiang Zhang

Researchers discovered that premature tracklet termination, rather than incorrect associations, causes 99.9% of identity switches in 3D Multi-Object Tracking (3DMOT) systems for autonomous driving. The proposed "Immortal Tracker" leverages a simple 3D Kalman filter to continuously predict and maintain tracklets even when objects are temporarily unobserved, reducing the mismatch ratio for vehicles on the Waymo Open Dataset to an unprecedented 0.0001 and achieving state-of-the-art overall tracking accuracy.

Key insights

  • Premature tracklet termination, rather than association errors, is the dominant cause of identity switches in modern 3DMOT systems, representing a fundamental flaw in traditional tracklet lifecycle management.
  • Maintaining tracklets of temporarily unobserved objects through simple trajectory prediction (e.g., with a Kalman filter) effectively mitigates identity switches without requiring complex, learned features.
  • The "immortal tracklet" concept is generalizable across various 3D object detectors and significantly improves tracking robustness for critical applications like autonomous driving.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.