Skip to content

Dataset · tomg-group-umd

16.9M detailed image captions

tomg-group-umd/pixelprose

Detailed synthetic descriptions of 16.9 million images for training and evaluating vision-language models.

What's inside

This consists of 16,896,214 image-caption pairs in English. The captions are synthesized by the Gemini 1.0 Pro Vision model and describe images in detail. The pairs are divided into CommonPool (6.5 million) and CC12M (9.1 million).

What you can do with it

  • Training an image captioning or vision-language model
  • Fine-tuning a text-to-image model using detailed captions
  • Evaluating VQA and image description
Tasks
Image → text, Text → image, Visual question answering
Languages
English
Size
10M–100M rows
Downloads this month
756
Likes
180

Models trained on it

How to load it

from datasets import load_dataset

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