# SpatialVID

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

- id: `spatialvid`
- kind: video for world models
- robot or device: in-the-wild video with camera pose and depth (class: sensor only)
- how collected: web video filtered and annotated
- size as stated: 21,000+ hours raw; 2.7M clips, 7,089 hours
- hours: 7,089 | hours only claimed: 21,000 | episodes: not stated
- year: 2025
- organisation: SpatialVID team
- licence: CC-BY-NC-SA-4.0 (class: non commercial)
- access: gated (class: gated)
- link (dataset): https://huggingface.co/datasets/SpatialVID/SpatialVID
- Hugging Face repository: `SpatialVID/SpatialVID`, 5,232 downloads in 30 days, https://huggingface.co/datasets/SpatialVID/SpatialVID
- note: Per-frame poses, depth, masks and motion instructions.
- files: https://huggingface.co/datasets/SpatialVID/SpatialVID/tree/main. It has no LeRobot metadata at its top level, so episodes are not marked there. If it keeps datasets in folders, this lists them: https://datasets.gurasees.com/episodes/SpatialVID/SpatialVID.md

Confirm the licence at the link before relying on it.