A Style-Based Generator Architecture for Generative Adversarial Networks

We propose an alternative generator architecture for generative adversarial networks, borrowing from style transfer literature. The new architecture leads to an automatically learned, unsupervised separation of high-level attributes (e.g., pose and identity when trained on human faces) and stochastic variation in the generated images (e.g., freckles, hair), and it enables intuitive, scale-specific control of the synthesis... (read more)

PDF Abstract CVPR 2019 PDF CVPR 2019 Abstract

Datasets


Introduced in the Paper:

FFHQ

Mentioned in the Paper:

LSUN CelebA-HQ Perceptual Similarity
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK RESULT BENCHMARK
Image Generation CelebA-HQ 1024x1024 StyleGAN FID 5.06 # 1
Image Generation FFHQ StyleGAN FID 4.43 # 4
Image Generation LSUN Bedroom 256 x 256 StyleGAN FID 2.65 # 1

Methods used in the Paper


METHOD TYPE
Convolution
Convolutions
Adaptive Instance Normalization
Normalization
R1 Regularization
Regularization
Leaky ReLU
Activation Functions
Dense Connections
Feedforward Networks
Feedforward Network
Feedforward Networks
WGAN-GP Loss
Loss Functions
Adam
Stochastic Optimization
StyleGAN
Generative Models