Automated In-the-Wild Data Collection for Continual AI Generated Image Detection

Thanasis Pantsios, Dimitrios Karageorgiou, Christos Koutlis, George Karantaidis, Olga Papadopoulou, Symeon Papadopoulos
Information Technology Institute, CERTH, Thessaloniki, Greece
The 5th ACM International Workshop on Multimedia AI against Disinformation (MAD '26), 2026

Abstract

The rapid advancement of generative artificial intelligence has introduced significant challenges for reliable AI-generated image detection. Existing detectors often suffer from performance degradation under distribution shifts and when encountering newly emerging generative models. In this work, we propose a data-centric continual adaptation framework for updating detectors in evolving environments. We introduce an automated, weakly supervised pipeline for constructing in-the-wild datasets through fact-check article retrieval. Combining this with generator-driven data within a continual learning framework enables robust adaptation and mitigates catastrophic forgetting. Extensive experiments on two state-of-the-art detectors, RINE and SPAI, show significant improvements of +9.14% and +8.01% in average accuracy, respectively.

BibTeX

@inproceedings{pantsios2026automated,
  title={Automated In-the-Wild Data Collection for Continual AI Generated Image Detection},
  author={Pantsios, Athanasios and Karageorgiou, Dimitrios and Koutlis, Christos and Karantaidis, George and Papadopoulou, Olga and Papadopoulos, Symeon},
  booktitle={The 5th ACM International Workshop on Multimedia AI against Disinformation (MAD '26)},
  year={2026},
  doi={10.1145/3810988.3812662}
}