Event

DrawSplat: Implicit Gaussian SLAM Guided by Atlanta World

1 reports · 1 sources

Fan Zhu and other authors submitted a preprint proposing DrawSplat, an implicit Gaussian SLAM framework guided by an Atlanta World prior for indoor reconstruction. The method first estimates camera poses, sparse geometry, and the gravity direction through joint point-line tracking, then uses the shared gravity component of the Atlanta World prior to infer structural planes from tracked lines and converts them into differentiable regularization terms on Gaussian centers and normals; the framework also combines an anchor-based implicit neural Gaussian representation with adaptive structural sampling. The authors report that experiments on multiple datasets show that DrawSplat achieves robust tracking, high-fidelity rendering, and geometrically consistent reconstruction in Gaussian-based SLAM systems, while maintaining a balance among quality, efficiency, and model size. These results are as stated in the paper and await independent verification.