CityGaussian
CityGaussian is a research-driven software series for high-quality large-scale scene reconstruction using 3D Gaussian Splatting. Developed by the Institute of Automation at the Chinese Academy of Sciences, it offers efficient real-time rendering and geometrically accurate modeling of complex urban environments. The package includes versions V1 and V2, with V2 being an ICLR 2025 publication focused on efficiency and precision. Key features include multi-GPU reconstruction with controllable memory costs, 2DGS-style mesh extraction, trajectory-aligned rendering, and mesh video generation with artifact removal. It supports joint pose and 3D Gaussian optimization to correct imperfect camera poses from foundation models like COLMAP or VGGT-X. The software integrates capabilities from Gaussian Lightning such as web viewing, MipSplatting, AbsGS, and StopThePop. It is designed for applications requiring scalable scene analysis, high-fidelity digital twins, and geometric performance evaluation. The project provides off