TerraFly-Forensics: A Dataset for Forensic Detection of Generated Map Images with Quality Assessment of Generative Models

The rapid advancement of generative models has significantly influenced various domains, including Remote Sensing (RS), through applications in map generation, navigation, and environmental monitoring. However, alongside these positive impacts, the misuse of generative models has raised critical concerns, such as adversarial attacks. Existing forensic detection studies rely on publicly available datasets but often lack a focus on map images, which are essential for navigation and critical decision-making tasks. To bridge this gap, we introduce TerraFly-Forensics, a comprehensive dataset comprising real and synthesized map images generated using advanced generative models. The real maps were downloaded using the TerraFly mapping system, and state-of-the-art generative models were applied to produce high-quality synthesized map images, accompanied by a detailed evaluation of their generative quality. We have conducted an experimental evaluation of several recent forensic detection methods on the TerraFly-Forensics dataset, providing insights into and contributing a new foundation for forensic detection in Remote Sensing.