Skip to content

Dataset · longvideobench

Long-video understanding benchmark up to 1 hour for LMMs

longvideobench/LongVideoBench

3,763 long captioned videos with questions to evaluate long-context multimodal models.

What's inside

Contains 3,763 internet videos up to one hour long with subtitles across various topics, paired with multiple-choice questions. Provided in text and table formats in English. Created to test large multimodal models (LMMs) on long video contexts.

What you can do with it

  • Evaluating long-video comprehension capabilities of multimodal models
  • Testing visual question answering (VQA) systems incorporating subtitles
Tasks
Multiple choice, Visual question answering
Languages
English
Size
1K–10K rows
Downloads this month
33.6K
Likes
64
Access
on request on Hugging Face

How to load it

from datasets import load_dataset

ds = load_dataset("longvideobench/LongVideoBench")