This database contains 100,000 synthetic images generated using StyleGAN. In this database, contrary to the 100K-GeneratedImages database, the StyleGAN network was trained using around 29,000 photos from 69 different models, considering face images from a more controlled scenario (e.g., with a flat background). Thus, no strange artifacts created by the StyleGAN are included in the background of the images.


State University of New York at Stony Brook

This manipulation creates entire non-existent face images. The table below summarises the main publicly available databases for research on detection of image manipulation techniques relying on entire face synthesis. Four different databases are of relevance here, all of them based on the same GAN architectures: ProGAN and StyleGAN. It is interesting to remark that each fake image may be characterised by a specific GAN fingerprint just like natural images are identified by a device-based fingerprint. In fact, these fingerprints seem to be dependent not only of the GAN architecture, but also of the different instances of it.

Manipulation Techniques and Public Databases of Entire Face Synthesis

Tolosana, R., Vera-Rodriguez, R., Fierrez, J., Morales, A., & Ortega-Garcia, J. (2020). Deepfakes and beyond: A survey of face manipulation and fake detection. Information Fusion, 64, 131-148.

DeepFakes and Beyond: A Survey of Face Manipulation and Fake Detection

Karras et al. released a set of 100,000 synthetic face images, named 100K-Generated-Images10. This database was generated using their proposed StyleGAN architecture, which was trained using the FFHQ dataset. StyleGAN is an improved version of their previous popular approach ProGAN, which introduced a new training methodology based on improving both generator and discriminator progressively. StyleGAN proposes an alternative generator architecture that leads to an automatically learned, unsupervised separation of high-level attributes (e.g., pose and identity when trained on human faces) and stochastic variation in the generated images (e.g., freckles, hair), and it enables intuitive, scale-specific control of the synthesis.

100K-Generated-Images10

100K-Faces

Regarding the entire face synthesis manipulation, the authors created 100,000 and 200,000 fake images through the pre-trained ProGAN and StyleGAN models, respectively.



Diverse Fake Face Dataset (DFFD)

This database comprises 250,000 and 80,000 synthetic face images created with StyleGAN and ProGAN, respectively. As an additional feature in comparison to previous databases, and in order to hinder fake detectors, in this database the fingerprints produced by the GAN architectures were removed through an approach named GANprintR (GAN fingerprint Removal), while keeping very realistic appearance. As a result of the GANprintR step, iFakeFaceDB presents a higher challenge for advanced fake detectors compared with the other databases.

Learn Before

Related