Ego-1K: 5.9 Million Multiview Egocentric Images from 12 Synchronous Cameras

Ego-1K is a collection of 956 short egocentric videos captured by Meta using a custom 12-camera synchronous rig for 3D video synthesis research. The dataset contains 491,000 frames and 5.9 million images documenting hand motions and dynamic scenes. Each video sequence lasts between 6.7 and 9.7 seconds, providing high-density multiview coverage from a VR headset perspective.

Use Cases

• Neural 3D video synthesis using the 12 synchronous camera perspectives
• Depth estimation training using multiview image sets
• Dynamic scene understanding of hand motions across 5.9 million images

Strengths

• 12 synchronous camera views per scene
• 5.9 million total images
• 491,000 high-resolution frames

Limitations

• Short video duration limited to under 10 seconds
• Narrow domain focus primarily on hand motions
• Geographic bias toward the United States region

Provenance

Source

Facebook (Meta)

Collection Method

Sensor-based capture using a custom 12-camera rig mounted on a VR headset

Freshness

Last updated March 2026.

Geography

United States