Semantic-Aware Temporal Adaptation for UAV Anti-UAV Tracking

Researchers proposed SATATrack, a new framework for UAV anti-UAV tracking that uses target descriptions as a stable semantic signal across frames.

The method adds contrastive learning against similar background regions and performs online distribution alignment at inference without updating model weights.

SATATrack achieved state-of-the-art results on the UAV-Anti-UAV benchmark and stayed competitive on related tracking tasks.

The approach addresses a hard low-altitude security problem where rapid viewpoint changes and distractors often break existing trackers.