Glow

E736216

Glow is a generative flow-based model architecture used for high-quality image and audio synthesis through invertible transformations.

Try in SPARQL Jump to: Statements Referenced by

Statements (48)

Predicate Object
instanceOf deep generative model
flow-based generative model architecture
normalizing flow model
appliedIn audio processing
computer vision
basedOn normalizing flows
canBeAppliedTo audio synthesis
speech modeling
comparedWith GANs NERFINISHED
VAEs
extends RealNVP NERFINISHED
field machine learning
hasAbbreviation Glow NERFINISHED
hasArchitectureComponent coupling layers
invertible 1x1 convolution layers
split operations
squeezing operations
hasAuthor Diederik P. Kingma NERFINISHED
Prafulla Dhariwal NERFINISHED
hasEvaluationMetric bits per dimension
log-likelihood
hasInfluenced subsequent normalizing flow models
hasKeyProperty efficient sampling
exact log-likelihood computation
invertible transformations
parallelizable architecture
tractable inference
hasKeyTechnique actnorm layers
affine coupling layers
invertible 1x1 convolutions
multi-scale architecture
hasLatentSpace continuous latent variables
hasProperty scalable to high-resolution images
supports conditional generation
hasPublicationYear 2018
hasTitle Glow: Generative Flow with Invertible 1x1 Convolutions NERFINISHED
hasTrainingObjective maximum likelihood estimation
implementedIn PyTorch NERFINISHED
TensorFlow NERFINISHED
improvesOver RealNVP NERFINISHED
publishedAt International Conference on Machine Learning NERFINISHED
subfield deep generative modeling
supports exact latent-variable inference
usedFor image editing
image generation
image synthesis
latent space interpolation
representation learning

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

WaveGlow basedOn Glow