Thanasis Pantsios, Dimitrios Karageorgiou, Christos Koutlis, George Karantaidis, Olga Papadopoulou, Symeon Papadopoulos
CERTH-ITI, Thessaloniki, Greece
Presented at MAD '26 – The 5th ACM International Workshop on Multimedia AI against Disinformation
This repository contains the official implementation and datasets for:
Automated In-the-Wild Data Collection for Continual AI-Generated Image Detection
The proposed framework introduces a continual adaptive pipeline for robust AI-generated image detection under evolving real-world conditions.
Figure 1: Overview of the proposed framework.
AIGenImages2026 is a dataset of images generated by recent text-to-image generative models. You can find more details in the paper.
- Dataset Download: Hugging Face Repository
- 5,439 generated images
- 19 recent generative models
- Chronologically organized generators
Example images from AIGenImages2026.
WildFC is an evolving in-the-wild dataset collected through an automated fact-check retrieval pipeline.
- Access Request: Hugging Face Repository
- 2,884 AI-generated images
- 2,298 segmented image samples
- Real-world fact-checked AI-generated content
Example images from WildFC.
Available pre-trained models of paper can be found here: RINE, SPAI
Guidelines for training and evaluation of our framework on RINE and SPAI can be found here and here
If you use our datasets or framework in your research, please cite the following paper:
@inproceedings{pantsios2026wildfc,
title={Automated In-the-Wild Data Collection for Continual AI Generated Image Detection},
author={Pantsios, Thanasis and Karageorgiou, Dimitrios and Koutlis, Christos and Karantaidis, George and Papadopoulou, Olga and Papadopoulos, Symeon},
booktitle={Proceedings of the 5th ACM International Workshop on Multimedia AI against Disinformation (MAD '26')},
year={2026}
}This work received funding from:
- AI-CODE (GA No. 101135437)
- ELIAS (GA No. 101120237)
If there are any questions, please feel free to contact:
Thanasis Pantsios
📧 [email protected]