iFakeFaceDB

[Paper]

iFakeFaceDB is face image dataset for the study of synthetic face manipulation detection, comprising about 87,000 synthetic face images generated by the Style-GAN model [1] and transformed with the GANprintR approach. All images were aligned and resized to the size of 224 x 224.

Sample Images

iFakeFaceDB Dataset Downloads

• downloading link

Source Code

The official implementation of the method used to generate this dataset can be found at https://github.com/joaocneves/gan_fingerprint_removal

License and Citation

The use of this software is RESTRICTED to non-commercial research and educational purposes.

@misc{1911.05351,
    title={{GANprintR: Improved Fakes and Evaluation of the State-of-the-Art in Face Manipulation Detection}},
    author={João C. Neves and Ruben Tolosana and Ruben Vera-Rodriguez and Vasco Lopes and Hugo Proença and Julian Fierrez},
    year={2019},
    eprint = {arXiv:1911.05351},
}

References

[1] Karras, T., Laine, S., & Aila, T. (2019). A style-based generator architecture for generative adversarial networks. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (pp. 4401-4410).