ARC-SORT: Adaptive, Robust, Compensated Multi-Object Tracking with Interacting Multiple Models
ARC-SORT is a training-free tracking method where a tracker's own uncertainty dictates its actions; it measures its uncertainty and uses that measurement to decide how to behave. It achieves better performance by using uncertainty to adapt matching regions, deferring identity commitments until evidence settles them, and merging tracks based purely on appearance during long gaps.