Tag
SAMOSA adapts SAM 2 for visual object tracking by incorporating motion prediction, semantic detection, and geometric constraints to improve robustness and generalization in complex scenarios with distractors, occlusion, and nonlinear motion.