facebookresearch/fast3r

Health v2 · maintenance only

Flags: no_releases archived no_license

Adoption not part of the score

1593 stars · 95 forks observed · 2026-08-28

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

Fast3R is the official PyTorch implementation of a CVPR 2025 model from Meta FAIR that reconstructs 3D scenes and estimates camera poses from up to 1500 images in a single Transformer forward pass. It generalizes the pairwise DUSt3R approach to many views in parallel, with a Gradio demo for uploading images or videos and visualizing reconstructions.

Use cases

• reconstruct a 3D scene from many unordered photos
• estimate camera poses from a video or image set in one forward pass
• generate point clouds from multi-view images without global alignment
• research baseline for multi-view 3D reconstruction
• visualize 3D reconstruction and confidence maps from uploaded video

When to choose

• you need fast multi-view 3D reconstruction from dozens to 1000+ images
• you want to avoid slow pairwise methods like DUSt3R with global alignment
• you need camera pose estimation as part of a vision pipeline
• you have a GPU and want a research-grade pretrained model

When to avoid

• you need a production-ready, commercially licensed 3D reconstruction tool (license is custom/non-standard)
• you have no GPU available
• you need real-time reconstruction on edge or mobile devices
• you need precise metric-scale reconstruction for surveying or CAD

Facets

library · maturity active

machine-learning computer-vision image-processing graphics simulation computer-vision machine-learning deep-learning graphics artificial-intelligence python windows 3d-reconstruction multi-view-stereo camera-pose-estimation transformer cvpr-2025 point-clouds gradio-demo research-code gpu linux macos

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/fast3r")