项目作者: Robinatp

项目描述 :
This is a repository of image stylization, include conditional instance normalization,conditional weight normalization,conditional stylized-params normalization,adaptive instance normalization and perceptural losses
高级语言: Python
项目地址: git://github.com/Robinatp/Tensorflow_image_stylization.git
创建时间: 2018-05-09T12:15:28Z
项目社区:https://github.com/Robinatp/Tensorflow_image_stylization

开源协议:

下载


Tensorflow_image_stylization

This is a repository of image stylization by Tensorflow implementation , include conditional instance normalization,conditional weight normalization,
conditional stylized-params normalization,adaptive instance normalization and perceptural losses

Reference

1,perceptural losses:fast-neural-style-tensorflow

Perceptual Losses for Real-Time Style Transfer and Super-Resolution, which first proposes an image transformation network
to transform input image into a pastiche outpur image and a loss network pretrained for image classification to define perceptual loss function

2,conditional instance normalization,conditional weight normalization:image_stylization

A Learned Representation for Artistic Style, which proposes conditional instance normalization,conditional weight normalization
to get multiple styles per module and combile the arbitrary styles by weight

3,conditional stylized-params normalization:arbitrary_image_stylization

Exploring the structure of a real-time, arbitrary neural artistic stylization network, which proposes a new style prediction
network that directly predicts an embedding vector S from an input style image for the style transfer network

4,adaptive instance normalization:Tensorflow-Style-Transfer-with-Adain

Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization, which propoces an AdaIn layer using to perform
style transfer in the feature sapce

All of method above have a fixed weight of a loss network pretrained for image classification in the train session,however in the inference there only a
style transfer network which build a encoder-decoder network except arbitrary_image_stylization which specially include style prediction network additionaly