LR-GAN: Layered Recursive Generative Adversarial Networks for Image Generation

Abstract

We present LR-GAN: an adversarial image generation model which takes scene structure and context into account. Unlike previous generative adversarial networks (GANs), the proposed GAN learns to generate image background and foregrounds separately and recursively, and stitch the foregrounds on the background in a contextually relevant manner to produce a complete natural image. For each foreground, the model learns to generate its appearance, shape and pose. The whole model is unsupervised, and is trained in an end-to-end manner with gradient descent methods. The experiments demonstrate that LR-GAN can generate more natural images with objects that are more human recognizable than DCGAN.

21 pages, 22 figures, published as a conference paper at ICLR 2017, code available on GitHub

Cited by in corpus (51)

• Progressive Growing of GANs for Improved Quality, Stability, and Variation
• Multimodal Unsupervised Image-to-Image Translation
• TransGAN: Two Pure Transformers Can Make One Strong GAN, and That Can Scale Up
• Ten Years of Generative Adversarial Nets (GANs): A survey of the state-of-the-art
• An Introduction to Image Synthesis with Generative Adversarial Nets
• RSGAN: Face Swapping and Editing using Face and Hair Representation in Latent Spaces
• Classification Accuracy Score for Conditional Generative Models
• BlockGAN: Learning 3D Object-aware Scene Representations from Unlabelled Images
• Unsupervised Object Segmentation by Redrawing
• Style Aggregated Network for Facial Landmark Detection
• Activation Maximization Generative Adversarial Nets
• AutoGAN: Neural Architecture Search for Generative Adversarial Networks
• Stacked Generative Adversarial Networks
• ST-GAN: Spatial Transformer Generative Adversarial Networks for Image Compositing
• Predicting Deeper into the Future of Semantic Segmentation
• Investigating Object Compositionality in Generative Adversarial Networks
• On the Evaluation of Conditional GANs
• Mask-aware Photorealistic Face Attribute Manipulation
• Convolutional Generation of Textured 3D Meshes
• MSG-GAN: Multi-Scale Gradients for Generative Adversarial Networks
• Creative Sketch Generation
• Characterizing Bias in Classifiers using Generative Models
• Emergence of Object Segmentation in Perturbed Generative Models
• AdversarialNAS: Adversarial Neural Architecture Search for GANs
• Composite Functional Gradient Learning of Generative Adversarial Models
• Compositional Transformers for Scene Generation
• A Survey and Taxonomy of Adversarial Neural Networks for Text-to-Image Synthesis
• Fine-grained Synthesis of Unrestricted Adversarial Examples
• Spatial Fusion GAN for Image Synthesis
• Understanding and Stabilizing GANs' Training Dynamics with Control Theory
• Towards Unsupervised Learning of Generative Models for 3D Controllable Image Synthesis
• Attention-GAN for Object Transfiguration in Wild Images
• MixNMatch: Multifactor Disentanglement and Encoding for Conditional Image Generation
• FineGAN: Unsupervised Hierarchical Disentanglement for Fine-Grained Object Generation and Discovery
• BIGPrior: Towards Decoupling Learned Prior Hallucination and Data Fidelity in Image Restoration
• How to make a pizza: Learning a compositional layer-based GAN model
• GaussiGAN: Controllable Image Synthesis with 3D Gaussians from Unposed Silhouettes
• Hierarchy Composition GAN for High-fidelity Image Synthesis
• Lipschitz Constrained GANs via Boundedness and Continuity
• Learning Segmentation Masks with the Independence Prior
• A Layer-Based Sequential Framework for Scene Generation with GANs
• Unsupervised Object Learning via Common Fate
• Towards a Neural Graphics Pipeline for Controllable Image Generation
• Defect-GAN: High-Fidelity Defect Synthesis for Automated Defect Inspection
• Quantifying point cloud realism through adversarially learned latent representations
• An Interpretable Generative Model for Handwritten Digit Image Synthesis
• Towards Disentangling Latent Space for Unsupervised Semantic Face Editing
• Conditional Transferring Features: Scaling GANs to Thousands of Classes with 30% Less High-quality Data for Training
• Dual Encoder-Decoder based Generative Adversarial Networks for Disentangled Facial Representation Learning
• Generating Object Stamps
• Unsupervised Learning of Depth and Depth-of-Field Effect from Natural Images with Aperture Rendering Generative Adversarial Networks