# TAP-Vid, TAPVid-3D and RoboTAP

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

- id: `tap-vid-tapvid-3d-and-robotap`
- kind: video for world models
- robot or device: real and synthetic video with point tracks; robot video in RoboTAP (class: sensor only)
- how collected: human annotation and synthetic ground truth
- size as stated: TAPVid-3D: 4,000+ real videos, 1M+ trajectories
- hours: not stated | hours only claimed: none | episodes: 1,000,000
- year: 2022
- organisation: Google DeepMind
- licence: Apache-2.0 (repo) (class: unclear)
- access: open (class: open)
- link (dataset): https://github.com/google-deepmind/tapnet
- note: Point tracking supervision used for robot motion priors.

Confirm the licence at the link before relying on it.