Skip to content

Dataset · HubLabs

First-person real manual labor video dataset

HubLabs/HubEgo-100K

Over 10,000 hours of real-world recordings of craftsmanship, cooking, and construction work with IMU sensor data.

What's inside

The first release of a large-scale egocentric dataset containing 10,124 hours of first-person video and IMU sensor readings from 12,227 participants (10 to 100 thousand clips). People unscripted recorded their actual manual labor on construction sites, in kitchens, workshops, salons, and at home. The data is packaged in LeRobot and MCAP formats.

What you can do with it

  • Training robots in fine motor skills and object manipulation
  • Training multimodal video and action understanding models (VLM)
Tasks
Robotics, Video classification, Video text → text
Languages
English
Size
1B–10B rows
Downloads this month
223.7K
Likes
93
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("HubLabs/HubEgo-100K")