PixelRNN

E203075

PixelRNN is a deep generative model that uses recurrent neural networks to sequentially model and generate images pixel by pixel.

All labels observed (4)

How this entity was disambiguated

Statements (47)

Predicate Object
instanceOf autoregressive model
deep generative model
image generation model
neural network architecture
application density estimation on images
image completion
image generation
architectureVariant Diagonal BiLSTM
PixelRNN self-linksurface differs
surface form: Multi-Scale PixelRNN

Row LSTM
codeAvailability open-source implementations by third parties
comparedWith PixelCNN
datasetUsed CIFAR-10
ImageNet patches
MNIST
developedAt DeepMind
surface form: Google DeepMind
factorizes joint pixel distribution into conditional distributions
generates images pixel by pixel
generationOrder raster-scan order
top-left to bottom-right
handles color images
grayscale images
natural images
improvesOn fully connected autoregressive models for images
inferenceType ancestral sampling
influenced PixelCNN
surface form: Gated PixelCNN

PixelCNN
surface form: PixelCNN++

subsequent autoregressive image models
inputType images
introducedInPaper PixelRNN self-linksurface differs
surface form: Pixel Recurrent Neural Networks
learningParadigm unsupervised learning
models images pixel by pixel
joint distribution of image pixels
modelType autoregressive density estimator
outputType images
paperAuthors Aaron van den Oord
Koray Kavukcuoglu
Nal Kalchbrenner
predicts each pixel conditioned on previously generated pixels
probabilisticModel yes
proposedIn 2016
relatedTo PixelCNN
trainingObjective log-likelihood maximization
maximum likelihood estimation
uses LSTM units
masked convolutions in some variants
recurrent neural networks

How these facts were elicited

Referenced by (5)

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

WaveNet relatedTo PixelRNN
PixelCNN relatedTo PixelRNN
PixelCNN comparedTo PixelRNN
this entity surface form: PixelRNN in terms of speed and performance
PixelRNN introducedInPaper PixelRNN self-linksurface differs
this entity surface form: Pixel Recurrent Neural Networks
PixelRNN architectureVariant PixelRNN self-linksurface differs
this entity surface form: Multi-Scale PixelRNN