# OpenHumanVid

From the catalogue. Listed by one researcher from one source page.

- id: `openhumanvid`
- kind: world model and video
- robot or device: none (human centric web video) (class: other)
- how collected: human centric clips from films, TV and web video with captions, skeletons and speech audio
- size as stated: 13.2 million filtered clips from 52.3 million raw clips; 16,700 hours released; 70,600 hours claimed
- hours: 16,700 | hours only claimed: 70,600 | episodes: not stated
- year: 2024
- organisation: Fudan University with Shanghai Jiao Tong University and Baidu
- licence: custom licence on application (class: custom terms)
- access: gated form (application reviewed) (class: gated)
- link (dataset): https://github.com/fudan-generative-vision/OpenHumanVid
- note: The paper's table gives the filtered high quality set as 13.2 million clips and 16.7K hours; the 70.6K figure is the raw pool.

Confirm the licence at the link before relying on it.