Reproduce “Iterative Scale-Up ExpansionIoU and Deep Features Association for Multi-Object Tracking in Sports” 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
Hsiang-Wei Huang, Cheng-Yen Yang, Jiacheng Sun, Pyong-Kun Kim, Kwang-Ju Kim, Kyoungoh Lee, Chung-I Huang, Jenq-Neng Hwang
Researchers from the University of Washington, ETRI, and National Center for High-Performance Computing developed Deep-EIoU, an online multi-object tracking method for sports videos that abandons the Kalman filter. The system leverages an iterative scale-up ExpansionIoU and deep appearance feature association to track athletes with irregular motions, achieving state-of-the-art performance on SportsMOT and SoccerNet-Tracking datasets.
Key insights
- Discarding the Kalman filter's linear motion assumptions is crucial for robustly tracking athletes whose movements are inherently irregular and non-linear.
- Intelligently expanding the bounding box search space via ExpansionIoU effectively mitigates issues arising from large, rapid object movements where traditional IoU-based methods fail due to lack of overlap.
- Iterative Scale-Up (ISU) for EIoU improves association reliability by dynamically adjusting the expansion scale, reducing sensitivity to initial hyperparameter choices and adapting to varied motion patterns.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.