Reproduce “Anti-UAV: A Large Multi-Modal Benchmark for UAV Tracking” 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

Nan Jiang, Kuiran Wang, Xiaoke Peng, Xuehui Yu, Qiang Wang, Junliang Xing, Guorong Li, Jian Zhao and 2 more

Researchers from the University of Chinese Academy of Sciences introduce Anti-UAV, a large multi-modal benchmark featuring unaligned RGB-T data for robust UAV tracking, alongside the Dual-Flow Semantic Consistency (DFSC) training strategy. The benchmark enables specialized research, and DFSC improves tracker performance, yielding gains on challenging attributes like thermal crossover.

Key insights

  • Unaligned multi-modal data represents a critical and realistic challenge in UAV tracking that requires specialized solutions for effective information fusion.
  • Domain-specific training on a large, diverse dataset like Anti-UAV yields substantial performance gains for UAV tracking compared to general object tracking benchmarks.
  • Separating class-level and instance-level semantic modulation during training effectively boosts tracker robustness and discriminative ability for single-category objects without runtime overhead.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.