GeoSynth: Contextually-Aware High-Resolution Satellite Image Synthesis

TL;DR

GeoSynth is a novel model that synthesizes high-resolution satellite images with controllable style and layout, using textual prompts and geographic data, enabling diverse and region-specific image generation.

Contribution

It introduces a new approach combining textual and geographic controls for satellite image synthesis, trained on a large, multi-source dataset for high-quality, diverse outputs.

Findings

Effective global style control via text and location

High-quality, diverse satellite image generation

Strong zero-shot generalization capabilities

Abstract

We present GeoSynth, a model for synthesizing satellite images with global style and image-driven layout control. The global style control is via textual prompts or geographic location. These enable the specification of scene semantics or regional appearance respectively, and can be used together. We train our model on a large dataset of paired satellite imagery, with automatically generated captions, and OpenStreetMap data. We evaluate various combinations of control inputs, including different types of layout controls. Results demonstrate that our model can generate diverse, high-quality images and exhibits excellent zero-shot generalization. The code and model checkpoints are available at https://github.com/mvrl/GeoSynth.