facebookresearch/House3D

Health v2 · maintenance only

Flags: archived

Adoption not part of the score

1200 stars · 185 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

House3D is a virtual 3D environment of over 45k fully annotated indoor scenes from the SUNCG dataset, built for training embodied AI agents. It provides fast rendering of RGB, depth, segmentation, and top-down map observations, making it suitable for large-scale reinforcement learning research.

Use cases

• train reinforcement learning agents for indoor navigation
• research embodied question answering in 3D environments
• simulate agents navigating to rooms by high-level concepts
• generate RGB, depth, and segmentation observations from 3D indoor scenes
• benchmark generalization of navigation models across environments

When to choose

• you need a large-scale, annotated 3D indoor environment for embodied AI or RL research
• you want fast rendering (thousands of FPS) for large-scale training
• you need multi-modal observations like depth, segmentation, and top-down maps

When to avoid

• you need outdoor or non-indoor environments
• you need actively maintained software with recent updates and support
• you need photorealistic rendering beyond SUNCG scene quality

Facets

library · maturity maintenance

simulation machine-learning reinforcement-learning computer-vision graphics artificial-intelligence reinforcement-learning computer-vision simulation cpp python 3d-environment embodied-ai indoor-scenes suncg visual-navigation reinforcement-learning-environment research linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/House3D")