Present and Future Generalization of Synthetic Image Detectors
It addresses the problem of detecting increasingly realistic synthetic images for security and verification applications, but is incremental in improving existing detector frameworks.
The paper systematically analyzes factors affecting generalization of synthetic image detectors and benchmarks state-of-the-art methods, finding no single detector universally effective and proposing workarounds to improve accuracy and robustness.
The continued release of increasingly realistic image generation models creates a demand for synthetic image detectors. To build effective detectors we must first understand how factors like data source diversity, training methodologies and image alterations affect their generalization capabilities. This work conducts a systematic analysis and uses its insights to develop practical guidelines for training robust synthetic image detectors. Model generalization capabilities are evaluated across different setups (e.g. scale, sources, transformations) including real-world deployment conditions. Through an extensive benchmarking of state-of-the-art detectors across diverse and recent datasets, we show that while current approaches excel in specific scenarios, no single detector achieves universal effectiveness. Critical flaws are identified in detectors, and workarounds are proposed to enable the deployment of real-world detector applications enhancing accuracy, reliability and robustness beyond the limitations of current systems.