BEGAN: Boundary Equilibrium Generative Adversarial Networks

By David Berthelot, Tom Schumm, Luke Metz. 2017

1,207 citations recorded by the source corpus.

We propose a new equilibrium enforcing method paired with a loss derived from the Wasserstein distance for training auto-encoder based Generative Adversarial Networks. This method balances the generator and discriminator during training. Additionally, it provides a new approximate convergence measure, fast and stable training and high visual quality. We also derive a way of controlling the trade-off between image diversity and visual quality. We focus on the image generation task, setting a new milestone in visual quality, even at higher resolutions. This is achieved while using a relatively simple model architecture and a standard training procedure.

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

Image Transformer (2018, background): “, 2015), motivating a large number of methods attempting to make their training more robust (Metz et al., 2016; Berthelot et al., 2017).”
Are GANs Created Equal? A Large-Scale Study (2017, methodology): “Various flavors of GANs have been recently proposed, both purely unsupervised [9, 1, 10, 5] as well as conditional [20, 21].”
Progressive Growing of GANs for Improved Quality, Stability, and Variation (2017, background): “GANs produce sharp images, albeit only in fairly small resolutions and with somewhat limited variation, and the training continues to be unstable despite recent progress (Salimans et al., 2016; Gulrajani et al., 2017; Berthelot et al., 2017; Kodali et al., 2017).”
Diagnosing and Enhancing VAE Models (2019, background): “7 Additionally, we present results from (Lucic et al., 2018) involving numerous competing GAN models, including MM GAN (Goodfellow et al., 2014), WGAN (Arjovsky et al., 2017), WGAN-GP (Gulrajani et al., 2017), NS GAN (Fedus et al., 2017), DRAGAN (Ko-dali et al., 2017), LS GAN (Mao et al., 2017) and ”
Improved Training of Wasserstein GANs (2017, neutral): “Nonlinearity (G) [ReLU, LeakyReLU, softplus(2x+2) 2 1, tanh] Nonlinearity (D) [ReLU, LeakyReLU, softplus(2x+2) 2 1, tanh] Depth (G) [4, 8, 12, 20] Depth (D) [4, 8, 12, 20] Batch norm (G) [True, False] Batch norm (D; layer norm for WGAN-GP) [True, False] Base filter count (G) [32, 64, 128] Base filte”
Negative Momentum for Improved Game Dynamics (2018, background): “Unfortunately, they are generally known to be difficult to train (Berthelot et al., 2017; Radford et al., 2015).”
Coulomb GANs: Provably Optimal Nash Equilibria via Potential Fields (2017, background): “Generative adversarial networks (GANs) [13] excel at constructing realistic images [36, 28, 24, 2, 4] and text [17].”
Which Training Methods for GANs do actually Converge? (2018, background): “For the ﬁrst class of stable equilibria, which we call energy solutions , the equilibrium discriminator forms an energy function for the true data distributions and might be a partial explanation for the success of autoencoder-based discrim-inators (Zhao et al., 2016; Berthelot et al., 2017).”
Demystifying MMD GANs (2018, background): “Note that the proof of Theorem 3 does not readily extend to generalized IPMs, and so does not tell us whether an unbiased estimator of the MMD GAN objective (14) can exist.”
GANs Trained by a Two Time-Scale Update Rule Converge to a Nash Equilibrium (2017, methodology): “The best FIDs that could be obtained with the original BEGANs and TTUR trained BEGANs are 28.55 and 26.19, respectively (see Table 1).”
Generating Adversarial Examples with Adversarial Networks (2018, background): “Most of Generative Adversarial Networks (GANs) (Goodfellow et al., 2014) have achieved visually appealing results in both image generation (Radford et al., 2015; Gulrajani et al., 2017; Berthelot et al., 2017) and manipulation (Zhu et al., 2016) settings.”
The Robust Manifold Defense: Adversarial Training using Generative Models (2017, methodology): “We show attacks for a gender classiﬁer trained on CelebA [30] and protected with the INC framework using a BEGAN [6] generator.”
Synthesizing Tabular Data using Generative Adversarial Networks (2018, background): “Generative Adversarial Networks Since GANs were ﬁrst proposed, many efforts have been made to speed up and stabilize the training process [42, 1, 18, 4], with application-based studies mainly focusing on generating images.”
RUDDER: Return Decomposition for Delayed Rewards (2018, background): “The entropy coefficient is replaced by Proportional Control [11, 10].”
Counterfactuals uncover the modular structure of deep generative models (2018, methodology): “com/Heumi/BEGAN-tensorflow) Boundary Equilibrium GAN (BEGAN) (Berthelot et al., 2017), which used to set a milestone in visual quality for higher resolution face images.”
Expectation Learning for Adaptive Crossmodal Stimuli Association (2018, background): “of the Wasserstein distance between the loss of the autoencoder reconstruction for real and generated data as a loss function [15].”
Adversarial Audio Synthesis (2018, background): “But despite their increasing fidelity at synthesizing images (Radford et al., 2016; Berthelot et al., 2017; Karras et al., 2018), GANs have yet to be demonstrated capable of synthesizing audio in an unsupervised setting.”