Skip to content

Dataset · facebook

Action100M: action video dataset

facebook/action100m-preview

Large English video dataset with temporal action segments and text descriptions.

What's inside

Action100M is compiled from 1.2 million instructional internet videos with a total duration of 14.6 years. From these, about 100 million temporally localized segments with open-vocabulary action labeling and detailed captions were extracted. The language is English, modalities are video and text; this is a preview version with a metadata size between 100k and 1M.

What you can do with it

  • Training video-text models
  • Action classification in video
  • Testing video understanding and world modeling models
Tasks
Video classification, Video text → text
Languages
English
Size
100K–1M rows
Downloads this month
1.2K
Likes
152

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("facebook/action100m-preview")