Generating High-Fidelity Images with Disentangled Adversarial VAEs and Structure-Aware Loss

Keyphrases

• Adversarial Training 25%
• Adversarial Variational Autoencoder 25%
• Blurriness 25%
• Blurry Image 25%
• CelebA Dataset 25%
• Code Vector 50%
• Decoder 25%
• Deep Generative Models 25%
• Encoder-decoder Architecture 25%
• Evidence Lower Bound 25%
• Feature Consistency 25%
• GAN Network 25%
• Generated Samples 25%
• High-frequency Characteristics 25%
• Latent Code 50%
• Latent Feature 25%
• Learning Representations 25%
• Loss Function 25%
• Manifold Structure 25%
• MNIST 25%
• Model Distribution 25%
• Perceptual Quality 25%
• Real Image 25%
• S-start 25%
• Sample Quality 25%
• Similarity Measure 25%
• Stable Training 25%
• Structure-aware 100%
• Target Distribution 25%
• Training Mechanism 25%
• Training Objectives 25%
• Variational Autoencoder 100%

Engineering

• Autoencoder 100%
• Blurriness 25%
• Code Vector 50%
• Distribution Model 25%
• Generated Sample 25%
• Generative Model 25%
• Global Structure 25%
• Local Frequency 25%
• Loss Function 25%
• Objective Function 50%
• Perceptual Quality 25%
• Similarities 25%
• Theoretical Basis 25%

Computer Science

• Adversarial Machine Learning 25%
• Autoencoder 100%
• Deep Generative Model 25%
• Frequency Feature 25%
• Objective Function 50%
• Perceptual Quality 25%
• Target Distribution 25%
• Theoretical Basis 25%
• Training Objective 25%