#24 · Primary category: Robotics & Embodied AI Simulation

MASt3R-SLAM

computer-vision cvpr2025 robotics slam

[CVPR 2025] MASt3R-SLAM: Real-Time Dense SLAM with 3D Reconstruction Priors

Project last updated:11/09/25

GitHub Stars

3.2K

Forks

375

Contributors

5

License

Other

Why we included this project

Robots, drones, and AR headsets all face the same basic problem: a camera moves, and the system has to figure out where it went and what the world around it looks like. MASt3R-SLAM tackles that in real time on a single GPU by combining a pretrained 3D reconstruction model with a lightweight motion tracker, producing camera trajectories plus metric point clouds and surfaces as you move. It is the reference implementation behind a CVPR 2025 paper, so the code leans research-grade: Python with CUDA components, scripts for standard benchmarks like TUM and EuRoC, and a live-demo path if you have a depth camera handy. That makes it a good starting point for researchers or engineers who want to see state-of-the-art dense SLAM running on their own data without assembling the whole stack themselves. Just note the non-commercial license before you plan a product around it.

Articles for this project

No articles for this project yet.

To suggest a topic or contribute an article, contact us.

Related projects in this category