Neural style transfer

网络风格迁移

作者:无用

本文通过学习吴恩达视频所做笔记

 目录

  简介

  可视化卷积层

  构建风格迁移网络

一、网络风格迁移简介

二、可视化卷积层

可视化深层卷积网络???这个问题我看过一篇文章,我会在后补上

  本网络共有5个卷积层,每个层的卷积核所检测的对象都不一致

  第一层

  一直重复直到第五层结束。层数往后,其检测的对象越明确。

三、构建风格迁移网络

  First we define the cost function.which means the degree of similary between two images.a and B is two superparamers

  The cost function contains two parts,include cost function of content and style,the next images is content cost function.

  The following content is style cost function.of course we understand the 'style' of an image.

  对图像风格的直觉

 

Thanks to you read

Parper : https://arxiv.org/ftp/arxiv/papers/1805/1805.10852.pdf

    https://arxiv.org/pdf/1508.06576v2.pdf

Examples : https://github.com/hzy46/fast-neural-style-tensorflow

 

posted @ 2018-08-16 13:28  weilongyitian  阅读(611)  评论(0编辑  收藏  举报