Skip to content

Dataset · ranjaykrishna

Images with dense object annotations

ranjaykrishna/visual_genome

108K images with objects, attributes, and pairwise relationships for vision and language tasks.

What's inside

Visual Genome contains over 108k images with dense annotations of objects, attributes, and relationships. On average, each image has 35 objects, 26 attributes, and 21 pairwise relationships. Annotation language is English; the data enables modeling objects and their relationships.

What you can do with it

  • Training object detection
  • Training VQA
  • Training image-to-text
Tasks
Image → text, Object detection, Visual question answering
Languages
English
Size
100K–1M rows
Downloads this month
2.9K
Likes
86

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("ranjaykrishna/visual_genome")