FaceID-GAN: Learning a Symmetry Three-Player GAN for Identity-Preserving Face Synthesis

By Yujun Shen, Ping Luo, Junjie Yan, Xiaogang Wang, Xiaoou Tang. 2018

180 citations recorded by the source corpus.

Face synthesis has achieved advanced development by using generative adversarial networks (GANs). Existing methods typically formulate GAN as a two-player game, where a discriminator distinguishes face images from the real and synthesized domains, while a generator reduces its discriminativeness by synthesizing a face of photorealistic quality. Their competition converges when the discriminator is unable to differentiate these two domains. Unlike two-player GANs, this work generates identity-preserving faces by proposing FaceID-GAN, which treats a classifier of face identity as the third player, competing with the generator by distinguishing the identities of the real and synthesized faces (see Fig.1). A stationary point is reached when the generator produces faces that have high quality as well as preserve identity. Instead of simply modeling the identity classifier as an additional discriminator, FaceID-GAN is formulated by satisfying information symmetry, which ensures that the real and synthesized images are projected into the same feature space. In other words, the identity classifier is used to extract identity features from both input (real) and output (synthesized) face images of the generator, substantially alleviating training difficulty of GAN. Extensive experiments show that FaceID-GAN is able to generate faces of arbitrary viewpoint while preserve identity, outperforming recent advanced approaches.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

Generative Adversarial Networks (2021, neutral): “FaceID GAN aims at synthesizing high quality images with different facial expressions or poses [8].”
Interpreting the Latent Space of GANs for Semantic Face Editing (2019, methodology): “To achieve this goal, current methods required carefully designed loss functions [26, 10, 32], introduction of additional attribute labels or features [23, 36, 3, 34, 31], or special architectures [11, 30] to train new models.”
Deep Face Recognition: A Survey (2018, methodology): “Unlike previous methods that treat classifier as a spectator, FaceID-GAN [214] proposed a three-player GAN where the classifier cooperates together with the discriminator to compete with the generator from two different aspects, facial identity and image quality respectively.”
Detection of AI-Generated Synthetic Faces (2022, background): “More sophisticated modifications let to change the pose or the facial expression [49, 58].”
Face identity disentanglement via latent space mapping (2020, background): “Another line of identity-preserving works [Liu et al. 2018; Shen et al. 2018; Tian et al. 2018] train GAN-based Image-to-Image translation networks that preserve the identity of the input image, while a sub-set of other attributes are controlled by a different image.”
Disentangled and Controllable Face Image Generation via 3D Imitative-Contrastive Learning (2020, methodology): “CGAN [31] has been widely used in face image synthesis tasks especially identity-preserving generation [47, 2, 51, 3, 42].”
A survey on face data augmentation for the training of deep neural networks (2019, background): “Improvement of GANs by (a)three-player GAN [146], (b)style-based generator [156], and (c)pyramidal adversarial discriminator [157].”
InterFaceGAN: Interpreting the Disentangled Face Representation Learned by GANs (2020, methodology): “To achieve this goal, existing methods require carefully designed loss functions [33], [34], [35], introduction of additional attribute labels [7], [36], [37], [38], [39], or special architectures [40], [41] to train new models.”
The Creation and Detection of Deepfakes (2020, background): “In [25], the authors propose FaceID-GAN which can reenacts t at oblique poses and high resolution.”
Disentangling in Latent Space by Harnessing a Pretrained Generator (2020, background): “Another line of identity-preserving works [52, 39, 58] train GAN-based Image-to-Image translation networks that preserve the identity of the input image, while a subset of other attributes are controlled by a different image.”