Skip to content

Dataset · FINEBENCH

FineBench for video VQA

FINEBENCH/FineBench

A large-scale multimodal benchmark for evaluating action understanding in video.

What's inside

It contains English-language images and text, with a volume of 100K to 1M records (~200k questions). Built on AVA v2.2 annotations with a focus on human actions and object interactions.

What you can do with it

  • Evaluating video-language model vision capabilities
  • Benchmarking multimodal video systems
Tasks
Visual question answering, Video text → text
Languages
English
Size
100K–1M rows
Downloads this month
48.3K
Likes
1

How to load it

from datasets import load_dataset

ds = load_dataset("FINEBENCH/FineBench")