GANZoo:千奇百怪的生成對抗網絡,都在這裡了73個

允中編譯整理

量子位出品|公眾號QbitAI

自從Goodfellow2014年提出這個想法之後，生成對抗網絡（GAN）就成了深度學習領域內最火的一個概念，包括LeCun在內的許多學者都認為，GAN的出現將會大大推進AI向無監督學習發展的進程。

於是，研究GAN就成了學術圈裡的一股風潮，幾乎每周，都有關於GAN的全新論文發表。而學者們不僅熱衷於研究GAN，還熱衷於給自己研究的GAN起名，比如什麼3D-GAN、BEGAN、iGAN、SGAN……千奇百怪、應有盡有。

今天，量子位決定帶大家逛逛GANs的動物園（園長：AvinashHindupur），看看目前世界上到底存活著多少GAN。

GAN—GenerativeAdversarialNetworks

3D-GAN—LearningaProbabilisticLatentSpaceofObjectShapesvia3DGenerative-AdversarialModeling

AdaGAN—AdaGAN:BoostingGenerativeModels

AffGAN—AmortisedMAPInferenceforImageSuper-resolution

ALI—AdversariallyLearnedInference

AMGAN—GenerativeAdversarialNetswithLabeledDatabyActivationMaximization

AnoGAN—UnsupervisedAnomalyDetectionwithGenerativeAdversarialNetworkstoGuideMarkerDiscovery

ArtGAN—ArtGAN:ArtworkSynthesiswithConditionalCategorialGANs

b-GAN—b-GAN:UnifiedFrameworkofGenerativeAdversarialNetworks

BayesianGAN—DeepandHierarchicalImplicitModels

BEGAN—BEGAN:BoundaryEquilibriumGenerativeAdversarialNetworks

BiGAN—AdversarialFeatureLearning

BS-GAN—Boundary-SeekingGenerativeAdversarialNetworks

CGAN—TowardsDiverseandNaturalImageDescriptionsviaaConditionalGAN

CCGAN—Semi-SupervisedLearningwithContext-ConditionalGenerativeAdversarialNetworks

CatGAN—UnsupervisedandSemi-supervisedLearningwithCategoricalGenerativeAdversarialNetworks

CoGAN—CoupledGenerativeAdversarialNetworks

Context-RNN-GAN—ContextualRNN-GANsforAbstractReasoningDiagramGeneration

C-RNN-GAN—C-RNN-GAN:Continuousrecurrentneuralnetworkswithadversarialtraining

CVAE-GAN—CVAE-GAN:Fine-GrainedImageGenerationthroughAsymmetricTraining

CycleGAN—UnpairedImage-to-ImageTranslationusingCycle-ConsistentAdversarialNetworks

DTN—UnsupervisedCross-DomainImageGeneration

DCGAN—UnsupervisedRepresentationLearningwithDeepConvolutionalGenerativeAdversarialNetworks

DiscoGAN—LearningtoDiscoverCross-DomainRelationswithGenerativeAdversarialNetworks

DualGAN—DualGAN:UnsupervisedDualLearningforImage-to-ImageTranslation

EBGAN—Energy-basedGenerativeAdversarialNetwork

f-GAN—f-GAN:TrainingGenerativeNeuralSamplersusingVariationalDivergenceMinimization

GoGAN—GangofGANs:GenerativeAdversarialNetworkswithMaximumMarginRanking

GP-GAN—GP-GAN:TowardsRealisticHigh-ResolutionImageBlending

IAN—NeuralPhotoEditingwithIntrospectiveAdversarialNetworks

iGAN—GenerativeVisualManipulationontheNaturalImageManifold

IcGAN—InvertibleConditionalGANsforimageediting

ngUsingaConditionalGenerativeAdversarialNetwork

ImprovedGAN—ImprovedTechniquesforTrainingGANs

InfoGAN—InfoGAN:InterpretableRepresentationLearningbyInformationMaximizingGenerativeAdversarialNets

LR-GAN—LR-GAN:LayeredRecursiveGenerativeAdversarialNetworksforImageGeneration

LSGAN—LeastSquaresGenerativeAdversarialNetworks

LS-GAN—Loss-SensitiveGenerativeAdversarialNetworksonLipschitzDensities

MGAN—PrecomputedReal-TimeTextureSynthesiswithMarkovianGenerativeAdversarialNetworks

MAGAN—MAGAN:MarginAdaptationforGenerativeAdversarialNetworks

MalGAN—GeneratingAdversarialMalwareExamplesforBlack-BoxAttacksBasedonGAN

MARTA-GAN—DeepUnsupervisedRepresentationLearningforRemoteSensingImages

McGAN—McGan:MeanandCovarianceFeatureMatchingGAN

MedGAN—GeneratingMulti-labelDiscreteElectronicHealthRecordsusingGenerativeAdversarialNetworks

MIX+GAN—GeneralizationandEquilibriuminGenerativeAdversarialNets(GANs

MPM-GAN—MessagePassingMulti-AgentGANs

MV-BiGAN—Multi-viewGenerativeAdversarialNetworks

pix2pix—Image-to-ImageTranslationwithConditionalAdversarialNetworks

PPGN—Plug&PlayGenerativeNetworks:ConditionalIterativeGenerationofImagesinLatentSpace

PrGAN—3DShapeInductionfrom2DViewsofMultipleObjects

—TextureSynthesiswithSpatialGenerativeAdversarialNetworks

SAD-GAN—SAD-GAN:SyntheticAutonomousDrivingusingGenerativeAdversarialNetworks

SalGAN—SalGAN:VisualSaliencyPredictionwithGenerativeAdversarialNetworks

SEGAN—SEGAN:SpeechEnhancementGenerativeAdversarialNetwork

SeqGAN—SeqGAN:SequenceGenerativeAdversarialNetswithPolicyGradient

SketchGAN—AdversarialTrainingForSketchRetrieval

SL-GAN—Semi-LatentGAN:Learningtogenerateandmodifyfacialimagesfromattributes

SRGAN—Photo-RealisticSingleImageSuper-ResolutionUsingaGenerativeAdversarialNetwork

SGAN—GenerativeImageModelingusingStyleandStructureAdversarialNetworks

SSL-GAN

StackGAN—StackGAN:TexttoPhoto-realisticImageSynthesiswithStackedGenerativeAdversarialNetworks

TGAN—TemporalGenerativeAdversarialNets

TAC-GAN—TAC-GAN—TextConditionedAuxiliaryClassifierGenerativeAdversarialNetwork

TP-GAN—BeyondFaceRotation:GlobalandLocalPerceptionGANforPhotorealisticandIdentityPreservingFrontalViewSynthesis

Triple-GAN—TripleGenerativeAdversarialNets

VGAN—GenerativeAdversarialNetworksasVariationalTrainingofEnergyBasedModels

VAE-GAN—Autoencodingbeyondpixelsusingalearnedsimilaritymetric

ViGAN—ImageGenerationandEditingwithVariationalInfoGenerativeAdversarialNetworks

WGAN—WassersteinGAN

WGAN-GP—ImprovedTrainingofWassersteinGANs

WaterGAN—WaterGAN:UnsupervisedGenerativeNetworktoEnableReal-timeColorCorrectionofMonocularUnderwaterImages

招聘