Researchers have developed IMPACT-HOI, a novel framework designed to improve the annotation of egocentric videos for human-object interactions. This mixed-initiative system constructs structured event graphs by incrementally resolving partially specified event states, guided by a controller that balances direct queries, suggestions, and completions. A user study demonstrated that IMPACT-HOI reduced manual annotation actions by 13.5% and achieved a high event match rate, while ensuring the preservation of human-confirmed decisions. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Improves efficiency and accuracy in creating structured datasets for robot manipulation training.
RANK_REASON This is a research paper describing a new framework for video annotation. [lever_c_demoted from research: ic=1 ai=1.0]