STATrack: Spatio-temporal adaptive transformer with consistency-aware memory distillation for visual tracking

Visual object tracking poses significant challenges due to complex variations in appearance, occlusions, and diverse motion patterns in real-world scenarios. This research presents STATrack, a novel tracking framework based on the Transformer architecture, aimed at addressing these challenges through three key contributions: (1) the Adaptive Spatio-Temporal Consistency Attention (ASTCA) module, which employs parallel attention mechanisms to effectively align and integrate multi-scale features, enhancing the model’s adaptability to appearance changes; (2) the Spatio-Temporal Memory Distillation Network (STMDN), which efficiently manages dynamic memory to retain and refine target-specific information across frames; and (3) the Spatio-Temporal Consistency (STC) loss, which enforces temporal coherence, significantly reducing tracking jitter and improving trajectory stability. Comprehensive experiments across seven challenging benchmarks demonstrate that STATrack achieves state-of-the-art performance, with 76.4% AO on GOT-10k, 72.7% AUC on LaSOT, and 84.7% AUC on TrackingNet, while maintaining real-time processing efficiency at 38 FPS. These results highlight the effectiveness of STATrack in enhancing tracking robustness and accuracy, establishing it as a promising solution for practical applications in visual tracking.