Skip to content

Dataset · facebook

Winoground: compositional reasoning

facebook/winoground

Image and caption pairs with identical words in different orders to test vision-language models.

What's inside

Winoground is a small English benchmark for vision-language compositional reasoning. It provides two images and two captions with identical words in different orders to test correct matching of images and texts. Data was manually collected and annotated by experts, with fewer than 1,000 examples.

What you can do with it

  • Evaluating compositional reasoning in VLMs
  • Image-text matching tests
  • Checking word-order sensitivity
Tasks
Image → text, Text → image, Image classification
Languages
English
Size
up to 1K rows
Downloads this month
2.3K
Likes
124
Access
on request on Hugging Face

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("facebook/winoground")