Monocular SLAM | Jinwoo Jeon

GaussianFlow SLAM: Monocular Gaussian Splatting SLAM Guided by GaussianFlow

Mon, 20 Apr 2026 00:00:00 +0000

We propose GaussianFlow SLAM, a monocular 3D Gaussian Splatting SLAM framework that leverages dense optical flow as a geometry-aware cue. By encouraging the projected motion of Gaussians, termed GaussianFlow, to align with optical flow, our method injects consistent structural supervision into both map reconstruction and camera pose estimation, even without depth measurements. We further introduce normalized error-based densification and pruning to refine inactive and unstable Gaussians, achieving superior rendering quality and tracking accuracy over state-of-the-art methods.

AIM-SLAM: Dense Monocular SLAM via Adaptive and Informative Multi-View Keyframe Prioritization with Foundation Model

Thu, 05 Mar 2026 00:00:00 +0000

We propose AIM-SLAM, a dense monocular SLAM framework that leverages dense pointmap predictions from a geometric foundation model (VGGT) and adaptively prioritizes informative multi-view keyframes. Our SIGMA module retrieves a candidate keyframe set via voxel overlap and information gain and adaptively determines its size, while a joint multi-view Sim(3) optimization enforces consistent alignment across the selected views, achieving state-of-the-art pose estimation and accurate dense reconstruction with ROS integration.