Reproduce “Multi-State Tracker: Enhancing Efficient Object Tracking via Multi-State Specialization and Interaction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shilei Wang, Gong Cheng, Pujian Lai, Dong Gao, Junwei Han
The Multi-State Tracker (MST) framework improves efficient visual object tracking by introducing multi-state specialization and interaction, powered by a novel Hidden State Adaptation-based State Space Duality (HSA-SSD) module. It achieves state-of-the-art performance on seven benchmarks, including a 4.5% AO increase on GOT-10K, while maintaining high inference speeds of 200 FPS on GPU and 36 FPS on an edge device.
Key insights
- Employing multiple specialized and interactively aggregated state representations offers a more comprehensive understanding of the target, enhancing robustness beyond single-layer feature methods.
- The Hidden State Adaptation-based State Space Duality (HSA-SSD) module is highly effective for refining and interacting diverse state features, enabling accuracy gains without compromising inference speed.
- The strategic integration of multi-state processing with an optimized State Space Duality architecture bridges the gap between high tracking performance and real-time efficiency on edge devices.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.