Reproduce “When Trackers Date Fish: A Benchmark and Framework for Underwater Multiple Fish 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
Weiran Li, Yeqiang Liu, Qiannan Guo, Yijie Wei, Hwa Liang Leo, Zhenbo Li
Researchers from China Agricultural University and the National University of Singapore introduce MFT25, a comprehensive dataset for underwater multiple fish tracking, and SU-T, a specialized tracking framework. This work establishes new performance benchmarks for fish tracking while quantitatively highlighting the unique challenges posed by aquatic environments and fish behavior.
Key insights
- Fish tracking requires domain-specific adaptations, as evidenced by the superior performance of SU-T's non-linear motion model (UKF) and specialized association metric (FishIoU) over generic alternatives.
- Fish exhibit significantly higher morphological similarity and more erratic, non-linear motion patterns compared to terrestrial objects, demanding tailored appearance and motion modeling.
- The integration of appearance-based re-identification models consistently improves identity preservation for fish, even with their high visual similarity.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.