SplaTAM

Why we included this project

SplaTAM is a Python/PyTorch implementation of a CVPR 2024 method that couples camera tracking with dense scene reconstruction through 3D Gaussian splatting, so a single RGB-D stream becomes a detailed map while the camera moves. For robotics and AR/VR engineers, that means one codebase that both localizes the sensor and builds the scene, without wiring together separate pieces. Researchers get a benchmarked baseline too: the repo runs against standard SLAM datasets, so reconstruction quality and tracking accuracy can be compared out of the box. It is clearly research software rather than a drop-in library, and you will likely adapt the code to your own sensors and scenes, but as a readable reference for how modern Gaussian-based SLAM.