Request access to Ego500

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Ego500 is released under CC BY-NC 4.0 (non-commercial use with attribution). For commercial licensing, contact raj@humanarchive.ai. Please complete the fields below; access requests are reviewed by the Human Archive team.

Log in or Sign Up to review the conditions and access this dataset content.

Ego500: The Most Densely Annotated Open-Source Egocentric Action Dataset

[DRAFT CARD — data upload pending. See the release post: https://ego500.humanarchive.ai]

Ego500 is 500 hours of in-the-wild first-person video of real-world work across 60+ environments (23 with more than four hours of footage), carrying over 310,000 structured action annotations.

Every second carries dense step annotations: what is being done, which phase of the task it belongs to, what each hand is holding, and which task-relevant objects are visible. Hands are engaged in 92% of steps. The dataset spans 390 unique task-environment combinations.

Annotation schema

Each video is annotated as a sequence of steps:

{
  "step": 1,
  "start_time_seconds": 0.0,
  "end_time_seconds": 3.0,
  "description": "Pick up the cloth from the counter with the right hand.",
  "task_phase": "task",
  "left_hand": "empty",
  "right_hand": "cloth",
  "objects_visible": ["cloth", "counter", "sink"]
}
  • Temporal segment boundaries
  • Imperative action descriptions
  • Task phases (task vs. task-related work)
  • Per-hand state, annotated independently, with "empty" as a first-class value
  • Visible objects drawn from a per-video curated inventory

Quality

Every released label is a model draft verified and corrected by a trained annotator, then approved by an independent reviewer. Annotators complete a qualification protocol before working on production footage (0.763 inter-annotator agreement on the qualification pilot).

Every video is anonymized with EgoBlur Gen 2 before publication.

License

CC BY-NC 4.0: free for research and other non-commercial use with attribution. For commercial licensing, contact raj@humanarchive.ai.

Contact

Human Archive — https://humanarchive.ai

Downloads last month
544