City-Mesh3R: Simulation-Ready City-Scale 3D Mesh Reconstruction from Multi-View Images

City-Mesh3R introduces a scalable framework for reconstructing high-fidelity 3D city-scale meshes directly from unordered image collections using a divide-and-conquer strategy. The method addresses limitations of existing NeRF and Gaussian Splatting approaches by producing watertight, simulation-ready meshes suitable for large urban scenes without prohibitive computational overhead.

City-Mesh3R represents a meaningful advancement in 3D reconstruction technology by tackling the scalability problem that has constrained previous approaches. Traditional methods struggle with city-scale projects because they either fail to produce complete geometry suitable for simulation or demand computational resources that grow prohibitively with scene size. This new framework shifts the paradigm by implementing topological image clustering and distributed processing, enabling reconstruction of arbitrarily large urban environments.

The technical innovation lies in the divide-and-conquer architecture that bypasses exhaustive image feature matching—a computational bottleneck in existing systems. By partitioning the sparse city map spatially and performing geometry-aware camera selection on independent clusters, the method achieves both efficiency and quality. The subsequent surface refinement through curvature-aware adaptive vertex density remeshing ensures regular, simulation-ready geometry that captures fine details while maintaining computational feasibility.

For the 3D modeling and simulation industries, this development opens practical pathways to urban-scale digital twin creation. Urban planning, autonomous vehicle simulation, architectural visualization, and game development all benefit from efficient, high-fidelity city reconstruction. The watertight mesh output is particularly valuable for physics-based simulation applications that require manifold geometry.

Looking forward, the technology's real-world validation will depend on performance benchmarks across diverse urban environments and integration with existing simulation pipelines. Key questions remain regarding reconstruction accuracy for complex architectural details, processing time for massive datasets, and how the method handles seasonal variations or dynamic urban elements. Industrial adoption will likely accelerate if the framework proves robust across varied geographic and climatic conditions.

• →City-Mesh3R achieves city-scale 3D reconstruction through divide-and-conquer spatial partitioning rather than exhaustive global processing
• →The framework produces watertight, simulation-ready meshes with regular geometry suitable for physics-based applications
• →Distributed end-to-end image-to-mesh processing eliminates need for resource-intensive feature matching across entire datasets
• →Curvature-aware adaptive remeshing balances computational efficiency with preservation of fine surface details
• →Technology enables practical digital twin creation for urban planning, autonomous systems, and visualization industries