OmniHuMo: Over 3.2 Million Motion-Capture Sequences from Internet Videos

Over 3.2 million high-quality motion-capture sequences sourced from internet videos, totaling more than 5,000 hours of data. The dataset was created by L-yiheng and was last updated on the Hugging Face platform in May 2026.

Use Cases

• Train human pose estimation models based on the large-scale motion-capture sequences.
• Develop generative models for human animation using the diverse motion data.
• Benchmark action recognition algorithms on the video-sourced motion data.
• Create synthetic training data for robotics and simulation based on real human motions.

Strengths

• Contains over 3.2 million individual motion-capture sequences.
• Totals more than 5,000 hours of motion data.
• Sourced from a diverse range of internet videos.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The specific file formats and data structure are unknown from the provided metadata.
• Data may reflect source bias inherent to videos collected from the internet.

Provenance

Source

Internet videos.

Collection Method

Motion-capture sequences extracted from video sources.

Freshness

Last updated 2026-05-31 13:55:45; freshness should be verified.