OmniHuman: Large-Scale Video Dataset for Human-Centric Tasks

OmniHuman is a large-scale video dataset for human-centric understanding and generation tasks. The released portion contains 200,000 videos and 20,000 unique identities. It was uploaded by author julia527 and last updated on May 8, 2026.

Use Cases

• Train human action recognition models based on the large-scale video collection.
• Develop video generation models for human-centric scenes based on the dataset's focus.
• Conduct research on multi-person interactions based on the 20% double-person videos.
• Perform identity-based analysis or re-identification tasks based on the 20,000 unique IDs.

Strengths

• Contains 200,000 released videos, providing substantial scale.
• Includes 20,000 unique identities, offering diversity for identity-related tasks.
• 20% of the release consists of double-person videos, enabling study of interactions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The description indicates only a small portion of the full data has been open-sourced.
• Row count and file formats are unknown, which may limit suitability assessment.

Provenance

Source

huggingface

Collection Method

Likely collected or compiled for human-centric AI research.

Time Range

null

Freshness

Last updated 2026-05-08 01:21:19; freshness should be verified.

Geography

null