Skip to content

Dataset · nvidia

10K 3D scenes for embodied AI

nvidia/SAGE-10k

10,000 realistic indoor 3D scenes with 565K objects for training and evaluating embodied AI.

What's inside

SAGE-10k is a large-scale collection of interactive indoor scenes created by the agentic pipeline from the SAGE paper. It contains 10,000 diverse scenes, 50 room types and styles, and 565,000 uniquely generated 3D objects, preserving small items. Metadata is in English.

What you can do with it

  • Training embodied AI policies in simulation
  • Evaluating text-to-3D scene generation
  • Benchmarking indoor navigation and interaction
Tasks
Text → 3d
Languages
English
Size
10K–100K rows
Downloads this month
117K
Likes
91

How to load it

from datasets import load_dataset

ds = load_dataset("nvidia/SAGE-10k")