book

Hands-On Image Generation with TensorFlow

by Soon Yau Cheong

December 2020

Intermediate to advanced

306 pages

6h 59m

English

Packt Publishing

Read now

Unlock full access

Why subscribe?ContributorsAbout the authorAbout the reviewersPackt is searching for authors like you
Who this book is forHow to use this bookWhat this book coversTo get the most out of this bookDownload the example code filesDownload the color imagesConventions usedGet in touchReviews
Technical requirements Understanding probabilitiesProbability distributionPrediction confidenceThe joint probability of pixelsGenerating faces with a probabilistic modelMean facesConditional probabilityProbabilistic generative modelsParametric modelingBuilding a PixelCNN model from scratchAutoregressive modelsPixelRNNBuilding a PixelCNN model with TensorFlow 2Summary
Technical requirementsLearning latent variables with autoencodersEncoderDecoderBuilding an autoencoderGenerating images from latent variablesVariational autoencodersGaussian distributionSampling latent variablesLoss functionGenerating faces with VAEsNetwork architectureFacial reconstructionGenerating new facesSampling trickControlling face attributesLatent space arithmeticFinding attribute vectorsFace editingSummary
Technical requirementsUnderstanding the fundamentals of GANsThe architecture of a GANValue functionsGAN training stepsBuilding a Deep Convolutional GAN (DCGAN)Architecture guidelinesBuilding a DCGAN for Fashion-MNISTTraining our DCGANChallenges in training GANsUninformative loss and metricsInstabilityVanishing gradientMode collapseBuilding a Wasserstein GANUnderstanding Wasserstein lossImplementing the 1-Lipschitz constraintRestructuring training stepsImplementing gradient penalty (WGAN-GP)Tweaking WGAN-GP for CelebASummary
Technical requirementsConditional GANsImplementing a conditional DCGANVariants of cGANImage translation with pix2pixDiscarding random noiseU-Net as a generatorLoss functionsImplementing a PatchGAN discriminatorTraining pix2pixUnpaired image translation with CycleGANUnpaired datasetCycle consistency lossBuilding CycleGAN modelsAnalysis of CycleGANDiversifying translation with BicyleGANUnderstanding architectureImplementing BicycleGANSummary
Technical requirementsNeural style transferExtracting features with VGGReconstructing contentReconstructing styles with the Gram matrixPerforming neural style transferImproving style transferFaster style transfer with a feed-forward networkA different style featureControlling styles with a normalization layerArbitrary style transfer in real timeImplementing adaptive instance normalizationStyle transfer network architectureArbitrary style transfer trainingIntroduction to style-based GANsMultimodal Unsupervised Image-to-Image Translation (MUNIT)Summary
Technical requirementsIntroduction to iGANUnderstanding manifoldImage editingSegmentation map-to-image translation with GauGANIntroduction to pix2pixHDSpatial Adaptive Normalization (SPADE)Implementing GauGANSummary

Technical requirementsProGAN overviewPixel normalizationIncreasing image variation with minibatch statisticsEqualized learning rateBuilding a ProGANBuilding the generator blocksBuilding the discriminator blocksProgressively growing the networkLoss functionGrowing painsImplementing StyleGANStyle-based generatorImplementing the mapping networkAdding noiseImplementing AdaINBuilding the generator blockTraining StyleGANSummary
Technical requirementsSpectral normalizationUnderstanding spectral normImplementing spectral normalizationSelf-attention modulesSelf-attention for computer visionImplementing a self-attention moduleBuilding a SAGANBuilding a SAGAN generatorConditional batch normalizationBuilding the discriminatorTraining the SAGANImplementing BigGANScaling GANsSkipping latent vectorsShared class embeddingOrthogonal regularizationSummary
Technical requirementsVideo synthesis overviewUnderstanding face video synthesisDeepFake overviewImplementing face image processingExtracting image from videoDetecting and localizing a faceDetecting facial landmarksAligning a faceFace warpingBuilding a DeepFake modelBuilding the encoderBuilding the decoderTraining the autoencodersSwapping facesImproving DeepFakes with GANsSummary
Reviewing GANsOptimizer and activation functionsAdversarial lossAuxiliary lossNormalizationRegularization Putting your skills into practiceDon't trust everything you readHow big is your GPU?Build your model using existing models Understand the model's limitationsImage processingImage inpaintingImage compressionImage super-resolutionText to imageVideo retargetingPose transferMotion transferNeural renderingSummary
Leave a review - let other readers know what you think

Content preview from Hands-On Image Generation with TensorFlow

Chapter 6: AI Painter

In this chapter, we are going to look at two generative adversarial networks (GANs) that could be used to generate and edit images interactively; they are iGAN and GauGAN . The iGAN (interactive GAN) was the first network to demonstrate how to use GANs for interactive image editing and transformation, back in 2016. As GANs were still in fancy at that time, the generated image quality was not impressive as that of today's networks, but the door was opened to the incorporation of GANs into mainstream image editing.

In this chapter, you will be introduced to the concepts behind iGANs and some websites that feature video demonstrations of them. There won't be any code in that section. Then, we will go over a more recent award-winning ...