Skip to content

Dataset · tomg-group-umd

Long video question-answering dataset

tomg-group-umd/cinepile

Around 300k training and 5k test questions for long-form video understanding.

What's inside

CinePile is a dataset and benchmark for long video question-answering. Collected via an LLM pipeline with human participation based on existing human data. Includes approximately 300k training and 5k test examples; the language is English, and the format is tables and text.

What you can do with it

  • Train video-text-to-text models
  • Evaluate long-form video understanding
  • LoRA for visual-question-answering
Tasks
Visual question answering, Video text → text
Languages
English
Size
100K–1M rows
Downloads this month
282
Likes
96
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("tomg-group-umd/cinepile")