MASt3R-SLAM
MASt3R-SLAM is a real-time dense Simultaneous Localization and Mapping (SLAM) system presented in the CVPR 2025 paper by Riku Murai, Eric Dexheimer, and Andrew J. Davison. It leverages 3D reconstruction priors from the MASt3R foundation model to achieve high-fidelity dense mapping and robust camera tracking. The software operates on standard monocular or RGB cameras, supporting live input from devices like Intel Realsense, as well as pre-recorded MP4 videos or folders of sequential RGB images. Key features include automatic camera calibration handling, support for known intrinsic parameters, and compatibility with major public benchmarks such as TUM-RGBD, 7-Scenes, EuRoC, and ETH3D. Built on PyTorch with specific CUDA version requirements, the system utilizes state-of-the-art deep learning models for metric depth estimation and geometric consistency. It is designed for researchers and developers working on robotics, augmented reality, and 3D perception, offering a turnkey solution for generating accurate dens