HumanTracker: Towards Comprehensive and Human-Aligned Motion Tracking Benchmark
The paper argues humanoid tracking benchmarks miss the contact failures people actually notice.
HumanTracker adds roughly 153 hours of optical motion trajectories from professional performers, grouped into four motion families with text labels for diagnosis. The authors also introduce HumanScore, a preference-aligned metric trained on 12K motion pairs. They report that it better matches human judgments than standard kinematic metrics, especially on support, contact timing, and foot-skating failures. ArXiv · AI/CL/LG's note
HumanTracker adds roughly 153 hours of optical motion trajectories from professional performers, grouped into four motion families with text labels for diagnosis. The authors also introduce HumanScore, a preference-aligned metric trained on 12K motion pairs. They report that it better matches human judgments than standard kinematic metrics, especially on support, contact timing, and foot-skating failures. ArXiv · AI/CL/LG's note
score 4