Segment Anything with Motion, Geometry, and Semantic Adaptation for Complex Nonlinear Visual Object Tracking

Researchers introduced SAMOSA, a new SAM 2-based visual object tracking framework for complex nonlinear motion.

It adds a lightweight nonlinear motion predictor plus semantic and geometric constraints to improve mask selection, memory filtering, and failure recovery.

The result is more stable tracking under occlusion, distractors, and other challenging real-world conditions.

Experiments show SAMOSA outperforms other SAM 2-based methods, supervised trackers, and anti-UAV benchmarks.