Skip to content

Dataset · maelic

Visual Genome 150 (VG150) in COCO-JSON

maelic/VG150-coco-format

Popular scene graph generation benchmark with 150 object classes and 50 relation types in standard COCO format.

What's inside

This dataset represents the standard VG150 split of the well-known Visual Genome corpus (100,000 to 1,000,000 records with images and English annotations), converted to COCO-JSON format. It includes the 150 most frequent object categories and 50 relation types between them, used in Scene Graph Generation research.

What you can do with it

  • Training object detectors and Scene Graph Generation models
  • Benchmarking visual relationship understanding systems on images
Tasks
Object detection
Languages
English
Size
100K–1M rows
Downloads this month
22K
Likes
0

How to load it

from datasets import load_dataset

ds = load_dataset("maelic/VG150-coco-format")