Reproduce “IMPACT-HOI: Supervisory Control for Onset-Anchored Partial HOI Event Construction” 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

Haoshen Zhang, Di Wen, Kunyu Peng, David Schneider, Zeyun Zhong, Alexander Jaus, Zdravko Marinov, Jiale Wei and 7 more

IMPACT-HOI introduces a mixed-initiative framework for collaboratively constructing high-quality Human-Object Interaction (HOI) event records from egocentric videos. This system reduces manual annotation actions by 13.5% per event and achieves a 46.67% complete event match rate, while robustly guaranteeing zero violations of human-confirmed data.

Key insights

  • Treating annotation as incremental partial event completion with explicit human-lock mechanisms allows for an efficient and high-quality human-AI collaboration in complex, multi-field data labeling tasks.
  • Jointly predicting and refining temporal onset and semantic labels through a statistics-guided cooperative refinement process is crucial for ensuring logical consistency across interdependent event fields.
  • A trust-calibrated supervisory controller that bounds machine initiative based on risk and confidence is essential for maintaining data integrity and fostering user trust, especially when human expertise must be preserved.

Summary by alphaXiv.

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