SAT-SKYLINES: 3D Building Generation from Satellite Imagery and Coarse Geometric Priors

2025-08-25

AI Summary

Existing satellite-image-based 3D building generation methods struggle to accurately reconstruct structural geometry from a single overhead view, while mainstream detail refinement approaches heavily rely on high-fidelity voxel inputs and cannot produce high-quality models from simple geometric priors (e.g., cuboids). To address this, we propose an end-to-end generative framework that jointly leverages satellite imagery and coarse geometric priors (e.g., bounding boxes), enabling flexible and lightweight structural control via noise-interpolation-driven geometric modeling transformations. We introduce Skylines-50K—the first large-scale, stylized 3D building dataset—and design a deep generative network that achieves coarse-to-fine geometric refinement without additional computational overhead. Experiments demonstrate strong generalization across diverse architectural forms, yielding 3D building models with accurate topology, rich surface details, and high fidelity.

Technology Category

• Machine Learning: Deep Neural Architectures and Foundation Models
• Computer Vision: Large Vision Models
• Natural Language Processing: Generation

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

We present SatSkylines, a 3D building generation approach that takes satellite imagery and coarse geometric priors. Without proper geometric guidance, existing image-based 3D generation methods struggle to recover accurate building structures from the top-down views of satellite images alone. On the other hand, 3D detailization methods tend to rely heavily on highly detailed voxel inputs and fail to produce satisfying results from simple priors such as cuboids. To address these issues, our key idea is to model the transformation from interpolated noisy coarse priors to detailed geometries, enabling flexible geometric control without additional computational cost. We have further developed Skylines-50K, a large-scale dataset of over 50,000 unique and stylized 3D building assets in order to support the generations of detailed building models. Extensive evaluations indicate the effectiveness of our model and strong generalization ability.

Problem

• Generating 3D buildings from satellite imagery alone
• Overcoming reliance on detailed voxel inputs for 3D reconstruction
• Modeling transformation from coarse geometric priors to detailed structures

Innovation

• Transforms noisy coarse priors to detailed geometries
• Enables flexible geometric control without extra cost
• Uses large-scale dataset for detailed building generation