ICP-3DGS: SfM-free 3D Gaussian Splatting for Large-scale Unbounded Scenes
This addresses the problem of enabling neural rendering methods like 3D Gaussian Splatting to work in outdoor and large-scale scenarios where obtaining camera poses is difficult, representing a novel method for a known bottleneck.
The paper tackles the challenge of 3D scene reconstruction without relying on preprocessed camera poses from structure-from-motion, proposing ICP-3DGS which integrates Iterative Closest Point with optimization-based refinement for accurate pose estimation and a voxel-based densification approach for large-scale scenes. Experiments show it outperforms existing methods in both camera pose estimation and novel view synthesis across indoor and outdoor scenes.
In recent years, neural rendering methods such as NeRFs and 3D Gaussian Splatting (3DGS) have made significant progress in scene reconstruction and novel view synthesis. However, they heavily rely on preprocessed camera poses and 3D structural priors from structure-from-motion (SfM), which are challenging to obtain in outdoor scenarios. To address this challenge, we propose to incorporate Iterative Closest Point (ICP) with optimization-based refinement to achieve accurate camera pose estimation under large camera movements. Additionally, we introduce a voxel-based scene densification approach to guide the reconstruction in large-scale scenes. Experiments demonstrate that our approach ICP-3DGS outperforms existing methods in both camera pose estimation and novel view synthesis across indoor and outdoor scenes of various scales. Source code is available at https://github.com/Chenhao-Z/ICP-3DGS.