CityDreamer is a research project on unbounded 3D city generation, and the related page is located at https://infinitescript.com/project/city-dreamer/。根据官方资料，其核心目标是从图像集合中学习并生成无边界的三维城市。
In terms of purpose, this project focuses on the relatively underexplored field of 3D city generation. The official description points out that in recent years a large amount of research has focused on 3D natural scene generation, but 3D city generation has been less explored because it is more challenging: humans are more sensitive to structural deformation in urban environments; at the same time, buildings, as objects of the same category, vary far more in appearance than relatively consistent objects such as trees in natural scenes, so generating 3D cities is more complex than generating 3D natural scenes.
In terms of main capabilities and design, CityDreamer is described as a compositional generative model designed specifically for unbounded 3D cities. It separates the generation of building instances from the generation of other background objects, such as roads, green spaces, and water, into different modules. In addition, the project constructs two datasets, OSM and GoogleEarth, containing a large number of real-world city images, to improve the realism of the generated 3D cities in layout and appearance.
In terms of usage, existing materials do not provide specific installation commands, code invocation methods, or operating steps for an online demo; relevant information is not currently available.
In terms of applicable scenarios, the official description indicates that it is aimed at research and applications in 3D city generation and can learn to generate unbounded 3D cities based on Google Earth imagery and OpenStreetMap; specific deployment scenarios and integration methods are not described in detail in the materials.
In terms of limitations and results, the official statement says that extensive experiments prove that CityDreamer outperforms existing state-of-the-art methods in generating a variety of realistic 3D cities, but it does not provide quantitative metrics, platform requirements, fees, or price information; these are not currently available.
Supplementary information: this project was completed by Haozhe Xie, Zhaoxi Chen, Fangzhou Hong, and Ziwei Liu of S-Lab at Nanyang Technological University; the page includes a Spotlight Video and mentions a Disqus comment system loading prompt. The product name CityDreamer, the URL, and the dataset names OSM and GoogleEarth are all retained as in the original text.
