Skip to content

Dataset · facebook

One million videos with text descriptions from Meta

facebook/PE-Video

A collection of 1 million videos with descriptions and 120,000 verified annotations for training video models.

What's inside

A large-scale dataset of 1 million diverse videos presented as part of the work on Perception Encoder. All videos have text descriptions, and over 120,000 clips are additionally provided with automatic and human-verified expert annotations.

What you can do with it

  • Train and evaluate multimodal encoders for video-text alignment
  • Fine-tune VLM models to generate detailed video descriptions
  • Create semantic search systems for video archives
Size
100K–1M rows
Downloads this month
16.2K
Likes
54

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("facebook/PE-Video")