facebookresearch/Replica-Dataset resource

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

1316 stars · 113 forks observed · 2026-08-28

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

The Replica Dataset is a collection of 18 high-quality 3D reconstructions of indoor spaces with dense geometry, HDR textures, glass/mirror surfaces, and semantic segmentation. The accompanying C++ SDK provides a viewer (ReplicaViewer) and headless renderer, plus exports for use in the AI Habitat framework for agent training.

Use cases

• benchmark SLAM and 3D reconstruction algorithms on indoor scenes
• train embodied AI agents in realistic indoor environments with Habitat
• render synthetic images with ground-truth geometry and semantics for computer vision research
• study semantic and instance segmentation of indoor spaces
• visualize high-fidelity 3D indoor reconstructions
• generate training data for neural rendering or scene understanding models

When to choose

• you need photorealistic indoor scenes with dense geometry and HDR textures
• you want ground-truth semantic segmentation and glass/mirror annotations
• you train or evaluate agents in AI Habitat
• you need headless rendering of synthetic views for ML research

When to avoid

• you need outdoor or large-scale scenes
• you want a lightweight dataset without a C++ SDK dependency
• your project requires a permissive open-source license (license is custom/non-standard)
• you need frequently updated or expanding scene collections

Facets

dataset · maturity maintenance

computer-vision machine-learning simulation graphics data-science computer-vision machine-learning simulation robotics graphics cpp windows indoor-scenes 3d-reconstruction semantic-segmentation habitat-simulator slam-benchmark renderer viewer linux macos desktop

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/Replica-Dataset")