UniDataPro/egocentric-video

Egocentric Dataset for Physical AI and Robotics

The dataset contains 239.3 hours of first-person videos for egocentric vision and egocentric tracking. Featuring multimodal data from egocentric views, it includes data annotations and motion capture for extracting 3d poses. It provides detailed 3d objects and 3d scenes using visual data from VR headsets to analyze hands motions and pose estimations.

By utilizing this dataset, researchers and developers can advance egocentric vision systems, train robotic manipulation policies, and benchmark object detection algorithms on real-world first-person footage with precise 3D pose annotations. - Get the data

Captured via VR headsets and 4 Zed cameras, the dataset integrates multimodal data including IMU signals and quaternion-based orientation from onboard sensors to support pose estimations and 3D reconstructions.

💵 Buy the Dataset: This is a limited preview of the data. To access the full dataset, please contact us at https://unidata.pro to discuss your requirements and pricing options.

The videos were captured via Pico VR headset and 4 Zed cameras across six subsets:

• Subset 1: 33 hours – Natural speed, hands appear only when necessary (Pico + Trackers)
• Subset 2: 84 hours – Slow motion, both hands always in frame, detailed kinematics (Pico + Trackers)
• Subset 3: 20 hours – Real speed, hands as needed, object transferring between surfaces (Pico + Trackers)
• Subset 4-6: 102.3 hours – Scripted object transfer tasks across 3 scenarios (Zed + Pico + Tracker)

Quaternion-based orientation from onboard sensor fusion supports 3D pose estimations and egocentric tracking across first-person perspectives.

Environments captured: Kitchen, bathroom, living room, and other home environments.

Activities included: Daily household actions, object transferring, hand-object interactions.

🌐 UniData provides high-quality datasets, content moderation, data collection and annotation for your AI/ML projects