Reproduce “TTNet: Real-time temporal and spatial video analysis of table tennis” 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

Roman Voeikov, Nikolay Falaleev, Ruslan Baikulov

TTNet is a multi-task deep learning system from OSAI designed for real-time table tennis video analysis. It simultaneously performs ball detection with an RMSE of 2 pixels, event spotting with 97.0% accuracy, and semantic segmentation with 97.5% IoU on the OpenTTGames dataset.

Key insights

  • A two-stage approach for small, fast-moving object detection can achieve high accuracy and efficiency by combining global context with local refinement.
  • Multi-task learning improves overall system performance and computational efficiency by sharing representations and leveraging complementary information between tasks.
  • Incorporating temporal information from video sequences is crucial for accurate event spotting in high-speed sports.

Summary by alphaXiv.

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