litlas
Log inCreate account

Improved Training of Wasserstein GANs

Authors: Ishaan Gulrajani, Faruk Ahmed, Martín Arjovsky, Vincent Dumoulin, Aaron C. Courville - Conference on Neural Information Processing Systems, NeurIPS 2017 cited by 11,032

Generative Adversarial Networks (GANs) are powerful generative models, but suffer from training instability. The recently proposed Wasserstein GAN (WGAN) makes progress toward stable training of GANs, but sometimes can still generate only low-quality samples or fail to converge. We find that these problems are often due to the use of weight clipping in WGAN to enforce a Lipschitz constraint on the critic, which can lead to undesired behavior. We propose an alternative to clipping weights: penalize the norm of gradient of the critic with respect to its input. Our proposed method performs better than standard WGAN and enables stable training of a wide variety of GAN architectures with almost no hyperparameter tuning, including 101-layer ResNets and language models over discrete data. We also achieve high quality generations on CIFAR-10 and LSUN bedrooms.

Improved Techniques forTraining GANsImproved Techniques for Training GANsf-GAN: TrainingGenerative Neural…f-GAN: Training Generative Neural Samplers using Variational Divergence MinimizationUnsupervisedRepresentation Learning…Unsupervised Representation Learning with Deep Convolutional Generative Adversarial NetworksMulti-class GenerativeAdversarial Networks…Multi-class Generative Adversarial Networks with the L2 Loss FunctionUnrolled GenerativeAdversarial NetworksUnrolled Generative Adversarial NetworksBEGAN: BoundaryEquilibrium Generative…BEGAN: Boundary Equilibrium Generative Adversarial NetworksWasserstein GANWasserstein GANBoundary-SeekingGenerative Adversarial…Boundary-Seeking Generative Adversarial NetworksStacked GenerativeAdversarial NetworksStacked Generative Adversarial NetworksAdversarially LearnedInferenceAdversarially Learned InferenceConditional ImageSynthesis With Auxiliar…Conditional Image Synthesis With Auxiliary Classifier GANsGAN（GenerativeAdversarial Nets）GAN（Generative Adversarial Nets）InfoVAE: InformationMaximizing Variational…InfoVAE: Information Maximizing Variational AutoencodersTowards the AutomaticAnime Characters…Towards the Automatic Anime Characters Creation with Generative Adversarial NetworksOn the effect of BatchNormalization and Weigh…On the effect of Batch Normalization and Weight Normalization in Generative Adversarial NetworksCoulomb GANs: ProvablyOptimal Nash Equilibria…Coulomb GANs: Provably Optimal Nash Equilibria via Potential FieldsEnhancing UnderwaterImagery Using Generativ…Enhancing Underwater Imagery Using Generative Adversarial NetworksGeometry Score: A MethodFor Comparing Generativ…Geometry Score: A Method For Comparing Generative Adversarial NetworksImproved Training withCurriculum GANsImproved Training with Curriculum GANsGenerative AdversarialNetworks (GANs)…Generative Adversarial Networks (GANs): Challenges, Solutions, and Future DirectionsMPCC: Matching Priorsand Conditionals for…MPCC: Matching Priors and Conditionals for ClusteringHSGAN: Reducing modecollapse in GANs by the…HSGAN: Reducing mode collapse in GANs by the latent code distance of homogeneous samplesEAGAN: EfficientTwo-Stage Evolutionary…EAGAN: Efficient Two-Stage Evolutionary Architecture Search for GANsPalette: Image-to-ImageDiffusion ModelsPalette: Image-to-Image Diffusion ModelsImproved Training ofWasserstein GANsImproved Training of Wasserstein GANsEarlier referencesFocus paperCiting papersOlderNewer

Click a node to pin it, click the empty canvas to go back to this paper, or hover to preview. Open a node’s page from its title.