Skip to content

Dataset · google

Detailed image descriptions

google/imageinwords

Hyper-detailed English image descriptions for training and evaluating image-to-text models.

What's inside

The dataset contains images and very detailed text descriptions collected via a human-in-the-loop scheme. The metadata indicates 1 to 10 thousand examples and English language. The card mentions evaluating readability, completeness, specificity, and hallucinations during fine-tuning.

What you can do with it

  • Fine-tune a vision-language model with detailed captioning descriptions
  • Evaluate hallucinations and description completeness in VLMs
  • Check the quality of text-to-image or object-detection pipelines
Tasks
Image → text, Text → image, Object detection
Languages
English
Size
1K–10K rows
Downloads this month
333
Likes
125

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("google/imageinwords")