EgoHome: open benchmark for dense sub-second egocentric video with a new Jaccard metric
AkshatS07 · x · 2026-11-11
Perceptron AI open-sourced EgoHome, a new dense benchmark for sub-second egocentric understanding, with all annotations human-reviewed and re-verified by the research team.
- Every household-chore video clip carries 4 annotation types: hand boxes, subtasks, hand actions, and hand captions
- Over 750 segments under 500ms
- Introduces "permissible subsegment Jaccard," a new metric accounting for subtask labels having differing and arbitrary granularity
- Data, metrics, and judge prompts are all open on Hugging Face (CC BY-NC-4.0, test split with 137 rows)
More from Research
- EmoHarness: scaffolding lifts small models' multimodal emotion recognition by up to 13.4 points — UIUC-CS · 2026-11-11
- Microsoft Research backdoors an AI agent by editing fewer than 10 dormant neurons, no training needed — DimitrisPapail · 2026-11-11
- UC Berkeley's SubWM watermark detects LLM distillation even after laundering attacks — sumanthd17 · 2026-11-11
- New paper ties grokking to Singular Learning Theory: generalization as manifold collapse — burny_tech · 2026-11-11
- New grokking study: generalization emerges from spontaneous structural collapse and compression — burny_tech · 2026-11-11
- TRIAD benchmark: no multimodal retrieval system leads across its six subsets — _reachsumit · 2026-11-11