VIGS-SLAM: visual-inertial 3D Gaussian Splatting SLAM hits SOTA on five datasets

rsasaki0109 · x · 2026-09-10

VIGS-SLAM (ECCV 2026, from ETH Zurich's cvg group with Marc Pollefeys) is a tightly coupled visual-inertial 3D Gaussian Splatting SLAM system that takes RGB frames plus IMU readings, tracks the camera trajectory robustly in real time, and reconstructs high-fidelity Gaussian maps, achieving state-of-the-art results across five challenging datasets. Code is open-sourced on GitHub with an iPhone capture pipeline, real-time demo, Docker setup, and pretrained models.

Original post →

More from Embodied

Embodied channel →