Generative Gaussian Splatting for Unbounded 3D City Generation

File Size (GB) # Points # Points (b) Used VRAM as #Points Increases (c) File Size as #Points Increases (a) Overview of the Proposed Method: GaussianCity PersistentNature (5.99 FPS) Gaussian Rasterizer Gaussian Attributes BEV-Point Decoder BEV-Point #Points: 20M Area: 1.5 km 2 S. LUT. P. Attr. … … … … … 3D-GS BEV-Point 3D-GS BEV-Point (d) Comparisons of Visuals and Runtimes on the GoogleEarth Dataset BEV Maps Figure 1. (a) Benefiting from the compact BEV-Point representation, GaussianCity can generate unbounded 3D cities using 3D Gaussian splatting (3D-GS). (b) As the number of points increases, VRAM usage during 3D-GS training rises significantly, whereas BEV-Point, acting as a compact representation, maintains a constant VRAM usage. (c) As the number of points increases, BEV-Point exhibits significantly lower growth in file storage compared to 3D-GS. (d) The proposed GaussianCity achieves not only superior generation quality but also the best efficiency in 3D city generation.