The relativistic discriminator: a key element missing from standard GAN

By Alexia Jolicoeur-Martineau. 2018

1,106 citations recorded by the source corpus.

In standard generative adversarial network (SGAN), the discriminator estimates the probability that the input data is real. The generator is trained to increase the probability that fake data is real. We argue that it should also simultaneously decrease the probability that real data is real because 1) this would account for a priori knowledge that half of the data in the mini-batch is fake, 2) this would be observed with divergence minimization, and 3) in optimal settings, SGAN would be equivalent to integral probability metric (IPM) GANs. We show that this property can be induced by using a relativistic discriminator which estimate the probability that the given real data is more realistic than a randomly sampled fake data. We also present a variant in which the discriminator estimate the probability that the given real data is more realistic than fake data, on average. We generalize both approaches to non-standard GAN loss functions and we refer to them respectively as Relativistic GANs (RGANs) and Relativistic average GANs (RaGANs). We show that IPM-based GANs are a subset of RGANs which use the identity function. Empirically, we observe that 1) RGANs and RaGANs are significantly more stable and generate higher quality data samples than their non-relativistic counterparts, 2) Standard RaGAN with gradient penalty generate data of better quality than WGAN-GP while only requiring a single discriminator update per generator update (reducing the time taken for reaching the state-of-the-art by 400%), and 3) RaGANs are able to generate plausible high resolutions images (256x256) from a very small sample (N=2011), while GAN and LSGAN cannot; these images are of significantly better quality than the ones generated by WGAN-GP and SGAN with spectral normalization.

Subfields: Computer Vision and Pattern Recognition, Generative Models

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 Generative Models

Citations in the corpus

Self-Attention Generative Adversarial Networks (2018, methodology): “, 2018; 2019), modifying the learning objectives and dynamics (Arjovsky et al., 2017; Salimans et al., 2018; Metz et al., 2017; Che et al., 2017; Zhao et al., 2017; Jolicoeur-Martineau, 2019), adding regularization methods (Gulrajani et al.”
Recent Progress on Generative Adversarial Networks (GANs): A Survey (2019, methodology): “Furthermore, a new network unit named Residual-in-Residual Dense Block (RRDB) based on the relativistic GAN [46] was introduced.”
AutoGAN: Neural Architecture Search for Generative Adversarial Networks (2019, neutral): “Referring to recent GAN benchmark studies [39, 32], we consider extending our search space with the attention/self-attention [64], the style-based generator [27], the relativistic discriminator [25] and various losses such as the Wasserstein loss [3], among others.”
ESRGAN: Enhanced Super-Resolution Generative Adversarial Networks (2018, methodology): “Relativistic discriminator [2] is developed not only to increase the probability that generated data are real, but also to simultaneously decrease the probability that real data are real.”
Adversarial score matching and improved sampling for image generation (2020, background): “, 2017; 2019; 2020) and have subsequently spawned a wealth of variants (Radford et al., 2015a; Miyato et al., 2018; Jolicoeur-Martineau, 2018; Zhang et al., 2019).”
Regularizing Generative Adversarial Networks under Limited Data (2021, background): “Generative adversarial networks (GANs) [2, 7, 13, 25, 29, 44, 68] aim to model the target distribution using adversarial learning.”
On Positive-Unlabeled Classification in GAN (2020, methodology): “For higher resolution data sets, SGAN failed to converge in 128x128 and 256x256 resolution datasets while LSGAN will be stuck at the early stage in 256x256 resolution dataset [8].”
Small-Object Detection in Remote Sensing Images with End-to-End Edge-Enhanced GAN and Object Detector Network (2020, methodology): “We use a relativistic discriminator [31] instead of a normal discriminator.”
Connections between Support Vector Machines, Wasserstein distance and gradient-penalty GANs (2019, methodology): “This provides an explanation as to why RpGANs do not perform very well unless using a gradient penalty (Jolicoeur-Martineau, 2018b).”
A Self-supervised Approach for Adversarial Robustness (2020, background): “The transfer-ability of different attacks is compared against the number of iterations in Figure 6.”
Enjoy Your Editing: Controllable GANs for Image Editing via Latent Space Navigation (2021, positive): “Generative Adversarial Networks (GANs) (Goodfellow et al., 2014) have signiﬁcantly improved realistic image generation in recent years (Brock et al., 2018; Jolicoeur-Martineau, 2019; Karras et al., 2017; 2019a;b; Park et al., 2019; Zhang et al., 2018).”
A Survey on Generative Adversarial Networks: Variants, Applications, and Training (2020, methodology): “eature matching [196], unrolled GAN [197], mini-batch discrimination [196], historical averaging [196], two time-scale update rule [198], hybrid model [72], self-attention GAN [199], relativistic GAN [203], label* Corresponding author, xilizju@zju.edu.cn 2 smoothing [204], sampling GAN [205], proper”
Cross-Domain Transferability of Adversarial Perturbations (2019, background): “Here, inspired by the generative adversarial network in [21], we propose a relativistic adversarial perturbation (RAP) generation approach that explicitly takes in to account the discriminator’s predictions on clean images.”