Abstract
Recently, 3D Gaussian Splatting (3DGS) has attracted widespread attention dueto its high-quality rendering, and ultra-fast training and rendering speed.However, due to the unstructured and irregular nature of Gaussian point clouds,it is difficult to guarantee geometric reconstruction accuracy and multi-viewconsistency simply by relying on image reconstruction loss. Although manystudies on surface reconstruction based on 3DGS have emerged recently, thequality of their meshes is generally unsatisfactory. To address this problem,we propose a fast planar-based Gaussian splatting reconstruction representation(PGSR) to achieve high-fidelity surface reconstruction while ensuringhigh-quality rendering. Specifically, we first introduce an unbiased depthrendering method, which directly renders the distance from the camera origin tothe Gaussian plane and the corresponding normal map based on the Gaussiandistribution of the point cloud, and divides the two to obtain the unbiaseddepth. We then introduce single-view geometric, multi-view photometric, andgeometric regularization to preserve global geometric accuracy. We also proposea camera exposure compensation model to cope with scenes with largeillumination variations. Experiments on indoor and outdoor scenes show that ourmethod achieves fast training and rendering while maintaining high-fidelityrendering and geometric reconstruction, outperforming 3DGS-based and NeRF-basedmethods.