FakeClue Dataset

FakeClue is a large-scale, multimodal dataset designed to aid in synthetic image detection and artifact explanation. It consists of over 100,000 images across many categories. Each image is paired with fine-grained artifact annotations in natural language, detailing the synthetic elements present within the image. The dataset includes images from open-source datasets like GenImage, FF++, and Chameleon, as well as specialized synthetic data generated for remote sensing and document images.

Citation

@misc{wen2025spotfakelargemultimodal,
      title={Spot the Fake: Large Multimodal Model-Based Synthetic Image Detection with Artifact Explanation}, 
      author={Siwei Wen and Junyan Ye and Peilin Feng and Hengrui Kang and Zichen Wen and Yize Chen and Jiang Wu and Wenjun Wu and Conghui He and Weijia Li},
      year={2025},
      eprint={2503.14905},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2503.14905}, 
}