Researchers Build Largest Ever Egocentric Dataset to Advance AI Perception
An international consortium led by the University of Bristol creates Ego4D, the world's largest egocentric video dataset, to help artificial intelligence see the world from a human perspective.
The University of Bristol joins an international consortium of 13 universities and Facebook AI to build Ego4D, the world's largest egocentric dataset. This massive project compiles over 3,025 hours of video footage captured by off-the-shelf, head-mounted cameras worn by 855 participants across nine different countries.
Bristol serves as the only UK representative in this global effort, contributing 270 hours of footage from 82 local participants. These videos capture daily activities from a first-person perspective, showing everything from playing musical instruments and gardening to assembling furniture and grooming pets.
This dataset allows artificial intelligence to learn from the exact visual data that humans process, paving the way for advanced augmented reality glasses that guide users through recipes or locate lost items. Alongside the video footage, the project provides five new challenging benchmarks that test AI models on understanding places, actions, and social interactions.