rmurai0610/MASt3R-SLAM

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

3167 stars · 374 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

MASt3R-SLAM is a real-time monocular dense SLAM system built on the MASt3R two-view 3D reconstruction prior, producing globally consistent camera poses and dense geometry at 15 FPS. It supports uncalibrated RGB video with a generic central camera model, plus camera tracking, pointmap fusion, loop closure, and global optimisation.

Use cases

• run real-time dense SLAM from a monocular RGB video
• reconstruct 3D geometry from uncalibrated camera footage
• track camera pose and build dense maps without known calibration
• evaluate SLAM on TUM and other benchmark datasets
• handle zooming or time-varying camera models in SLAM
• research dense SLAM with learned 3D reconstruction priors

When to choose

• you need real-time dense SLAM from a single RGB camera without calibration
• you want state-of-the-art monocular SLAM benchmark performance with known calibration
• your videos are in-the-wild with no fixed parametric camera model

When to avoid

• you need a production system with a permissive license (license is non-standard)
• you lack a CUDA-capable GPU
• you need stereo, LiDAR, or multi-camera SLAM
• you need a lightweight CPU-only solution

Facets

library · maturity active

computer-vision machine-learning simulation computer-vision robotics autonomous-vehicles python slam 3d-reconstruction monocular-camera real-time pointmap loop-closure cvpr2025 research-code linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="rmurai0610/MASt3R-SLAM")