What’s Out There
← Back to swiping
Images & videoFreeCompany

Kinetics-700

Google DeepMind (hosted by the Common Visual Data Foundation)

650,000 ten-second YouTube clips of people doing 700 things, from yoga to changing a tire.

Human-action video clips, each about 10 seconds long and labeled with one of 400, 600, or 700 action classes depending on version. It is the main benchmark and pretraining set for video recognition. Because the source videos live on YouTube, many links died over time, so CVDF now hosts copies of Kinetics-400 and Kinetics-700-2020.

Access

Cost
FreeFree download scripts for hosted copies.
License
License not statedNot stated on the hosting page; videos remain YouTube content owned by uploaders
Sign-up
No sign-up
How you get it
Bulk downloadMP4, CSV

What's in it

One row is
One roughly 10-second video clip with one action label
Covers
YouTube videos worldwide
Years
Kinetics-400 released 2017; Kinetics-700-2020 in 2020
Updated
One-time release
Size
Up to about 650,000 clips across 700 classes, with at least 700 clips per class
YouTube IDstart timeend timeaction labelsplit

Quality

4 of 5

The standard video benchmark with human-verified labels; versions overlap, and mixing them leaks test data.

  • Kinetics-400 validation clips largely appear in Kinetics-700 training data
  • Original YouTube links decay over time
  • Uploaders did not consent to research use

Connects to

  • AudioSet: Also built from 10-second YouTube segments, labeled for sound instead of action
  • Ego4D: First-person video of daily activity, complementing third-person action clips

Questions it could answer

  • Which actions are hardest to tell apart from a single frame?
  • What share of the original YouTube clips are still online?

Similar data sets

  • FineART

    40,543 robot episodes and 533,913 human-written subtask labels covering 1,718 hours.

    Images & videoScale AI and Hugging Face LeRobotFree

  • AgiBot World 2026

    Real-world robot demonstrations from homes and commercial spaces, collected on the AgiBot G2 platform.

    Images & videoAgiBotFree

  • Ego4D

    3,670 hours of daily life filmed from head-mounted cameras by 923 people in 9 countries.

    Images & videoEgo4D consortium (Meta AI and university partners)By application

  • AudioSet

    Two million 10-second YouTube clips labeled with 527 kinds of sound, from barking to bagpipes.

    Audio & musicGoogle Research (Sound Understanding group)Free

  • ABC-130k

    Billed as the largest open-source robot teleoperation dataset: bimanual trajectories on two-arm YAM stations.

    Images & videoXDOFFree

  • ImageNet

    The 14-million-photo set whose 2012 contest kicked off the deep-learning boom.

    Images & videoStanford University and Princeton University (ImageNet project)Free