Dataset Card for So-Fake-Set

Dataset Description

• Project Page: https://hzlsaber.github.io/projects/So-Fake/
• Repository: https://github.com/hzlsaber/So-Fake
• Point of Contact: Zhenglin Huang

Dataset Summary

We provide So-Fake-Set, A large-scale, diverse dataset tailored for social media image forgery detection!

Please check our website to explore more visual results.

Dataset Structure

• "image" (Image): Input images, including real, full_synthetic, and tampered images.

• "mask" (Image): Binary mask highlighting manipulated regions in tampered images.

• "label" (str): Classification category.

• "generator" (str): The generator/source model. For real images this field is None.

• "filename" (str): Original filename of the image.

• "split" (str): train and validation.

Licensing Information

This work is licensed under a Creative Commons Attribution 4.0 International License.

Citation Information

If you find this dataset useful, please consider citing our paper:

@misc{huang2025sofakebenchmarkingexplainingsocial,
      title={So-Fake: Benchmarking and Explaining Social Media Image Forgery Detection}, 
      author={Zhenglin Huang and Tianxiao Li and Xiangtai Li and Haiquan Wen and Yiwei He and Jiangning Zhang and Hao Fei and Xi Yang and Xiaowei Huang and Bei Peng and Guangliang Cheng},
      year={2025},
      eprint={2505.18660},
      archivePrefix={arXiv},
      url={https://arxiv.org/abs/2505.18660}, 
}