Sat2RealCity: Geometry-Aware and Appearance-Controllable 3D Urban Generation from Satellite Imagery

Jul 16, 2026

Upload by Patrick

3D urban generation from satellite imagery is essential for scalable digital twins and realistic simulation. Existing methods often depend on scene-level generation, which struggles with controllability, geographic alignment, and grounding realistic appearances using large 3D city assets. Sat2RealCity introduces a grounded framework that decomposes cities into geographically anchored building entities, reusing object-level 3D generative priors. BuildVerse3D dataset supports OSM-guided spatial grounding, appearance-guided controllable synthesis, and an MLLM semantic pipeline for consistent regional styles.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages15

Reading Time38 min

Introduction

Problem and motivation

Proposed framework overview

Dataset and key components

arXiv :2511 . 11470v2 [ cs .CV] 13 Jul 2026 Sat2RealCity: Geometry-Aware and Appearance-Controllable 3D Urban Generation from Satellite Imagery Xinliang Wang* , Yijie Kang* , Zhenyu Wu, and Yifeng Shi† KE Holdings Inc., Beijing, China {wangxinliang008, kangyijie001, wuzhenyu018, [shiyifeng003](shiyifeng003}@ke.com)[}](shiyifeng003}@ke.com)[@ke.com](shiyifeng003}@ke.com) [Abstract.](Abstract. 3D urban generation from satellite imagery is an important)[ 3D urban generation from satellite imagery is an important](Abstract. 3D urban generation from satellite imagery is an important)[ ](Abstract. 3D urban generation from satellite imagery is an important)[task for scalable digital twins and real-world simulation environments.](task for scalable digital twins and real-world simulation environments.)[ ](task for scalable digital twins and real-world simulation environments.)[Existing approaches primarily rely on scene-level generation paradigms](Existing approaches primarily rely on scene-level generation paradigms), which often require large-scale 3D city assets and struggle with controllability, geographic alignment, and realistic appearance grounding in real-world urban environments. To address these limitations, we present Sat2RealCity, a grounded urban generation framework that leverages object-level 3D generative priors for scalable city synthesis from satellite imagery. Our framework decomposes cities into geographically grounded building entities, enabling the reuse of pretrained object-level 3D generative priors while preserving real-world spatial structures. Supported by our constructed BuildVerse3D dataset, (1) we introduce an OpenStreetMap (OSM)-guided spatial grounding strategy to inject geospatial constraints into the 3D generation process; (2) we design an appearanceguided controllable generation mechanism for realistic architectural appearance and regional style consistency; and (3) we construct an MLLMpowered semantic pipeline for regional appearance understanding and semantic-aware appearance synthesis. Extensive experiments demonstrate that Sat2RealCity achieves strong geographic alignment, regional stylistic consistency, and plausible urban asset synthesis compared with existing urban generation and 3D asset generation approaches. Keywords: 3D Urban Generation · Satellite Imagery · Building-Entity 3D Generation 1 Introduction Generating large-scale 3D urban environments from satellite imagery is valuable for digital twins, urban visualization, and large-scale spatial content creation. Advances in dense visual prediction, object detection, monocular 3D perception, and transportation foundation models have improved the perception and understanding of complex urban environments [41,38,11,17,32] . Complementary to these perception-centric efforts, recent progress in 3D spatial understanding, reconstruction, and generative world modeling has advanced the modeling * Equal contribution. †Corresponding author. 2 X. Wang et al. Sat2RealCity Satellite Imagery Generated 3D Urban Fig. 1. We present Sat2RealCity, a framework for generating explicit 3D urban assets with controllable geometry and appearance from real-world satellite imagery. of coherent 3D environments [44,46,45,39,8,18,9] . Sat2RealCity focuses on a distinct but related problem: generating explicit, editable, and geographically grounded urban assets from overhead imagery. Nevertheless, existing urban generation approaches still face several practical limitations. One line of work [7, 43,21] focuses on scene-level 3D generation using dense representations such as NeRF [23], 3DGS [12], or heavy voxel grids. While these methods demonstrate promising large-scale urban synthesis capabilities, they often require large-scale 3D city assets and struggle with controllability, geographic alignment, and realistic appearance grounding in real-world urban environments. Another line of work [22,33,19,26,48] focuses on cross-view image synthesis fr

FAQs

Why is 3D urban generation from satellite imagery challenging with existing scene-level approaches?

Scene-level methods typically require large-scale 3D city assets and face difficulties in controllability, geographic alignment, and realistic appearance grounding in real-world urban settings.

How does Sat2RealCity achieve scalable and editable urban asset generation?

Sat2RealCity decomposes cities into geographically grounded building entities, enabling compositional generation that reuses object-level 3D generative priors while preserving real-world spatial structures.

What are the main components used during inference to improve alignment and appearance realism?

The method uses an OSM-based spatial priors strategy for geospatial footprints alignment, an appearance-guided controllable mechanism for regional style consistency, and an MLLM-powered semantic pipeline to derive semantic-aware appearance references.

Jul 16, 2026

Upload by Patrick