Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic Image
2017-10-12 19:56
701 查看
着色原理
网络结构为:
输入为灰度图像,输出为彩色图像.网络结构包括四个部分,Low-Level Feature Network用于提取图像的低层特征,提取低层特征后,通过一个Mid-Level Feature Network得到中间特征,另外通过Global Feature Network提取全局特征,将全局特征与中间特征融合,即Fusion layer,之后将融合特征输入Colorization Network得到图像的颜色特征,将颜色特征与输入图像(亮度图像)结合,得到最后的彩色图像.
网络具体参数如下:
特征融合(Fusion layer)计算为:
ymu,vid,yglobal为256维的特征向量,w为256×512的矩阵,yfusionu,v为256维的向量.
损失函数
同时将Global Feature Network特征输入到一个Classification Network,对数如图像进行分类,以优化Global Feature Network网络.这样,损失函数包括分类损失函数,以及彩色图像与真实图像的损失函数,计算如下:
效果对比
第一行为输入灰度图像,第二行为核心算法效果,第三行为不填加Global Feature Network的效果,第四行为添加Global Feature Network的效果.可见,在添加Global Feature Network特征后,图像颜色更丰富,饱和度更高.
网络结构为:
输入为灰度图像,输出为彩色图像.网络结构包括四个部分,Low-Level Feature Network用于提取图像的低层特征,提取低层特征后,通过一个Mid-Level Feature Network得到中间特征,另外通过Global Feature Network提取全局特征,将全局特征与中间特征融合,即Fusion layer,之后将融合特征输入Colorization Network得到图像的颜色特征,将颜色特征与输入图像(亮度图像)结合,得到最后的彩色图像.
网络具体参数如下:
特征融合(Fusion layer)计算为:
ymu,vid,yglobal为256维的特征向量,w为256×512的矩阵,yfusionu,v为256维的向量.
损失函数
同时将Global Feature Network特征输入到一个Classification Network,对数如图像进行分类,以优化Global Feature Network网络.这样,损失函数包括分类损失函数,以及彩色图像与真实图像的损失函数,计算如下:
效果对比
第一行为输入灰度图像,第二行为核心算法效果,第三行为不填加Global Feature Network的效果,第四行为添加Global Feature Network的效果.可见,在添加Global Feature Network特征后,图像颜色更丰富,饱和度更高.
相关文章推荐
- Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic Image
- 论文笔记之:Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic
- 论文笔记之:Let there be Color!: Joint End-to-end Learning of Global and Local Image Priors for Automatic
- 立体匹配——End-to-End Learning of Geometry and Context for Deep Stereo Regression
- End-to-End Learning of Deformable Mixture of Parts and Deep Convolutional Neural Networks for Human
- Let there be Color!: Automatic Image Colorization with Simultaneous Classification - sig - 2016
- End-to-End Learning of Geometry and Context for Deep Stereo Regression
- 论文阅读:End-to-End Learning of Deformable Mixture of Parts and Deep Convolutional Neural Networks for H
- BaiXiang——【arXi2015】An End-to-End Trainable Neural Network for Image-based Sequence Recognition and
- End-to-End Reinforcement Learning of Dialogue Agents for Information Access
- "Can I use" provides up-to-date browser support tables for support of front-end web technologies on desktop and mobile web browsers.
- 1607.CVPR-Joint Learning of Single-image and Cross-image Representations for Person ReID 论文笔记
- hbase启动报错hbase-default.xml file seems to be for and old version of HBase
- 论文阅读(Xiang Bai——【PAMI2017】An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application to Scene Text Recognition)
- 姿态检测整理--06-Associative Embedding: End-to-End Learning for Joint Detection and Grouping
- 论文笔记:An End-to-End Trainable Neural Network for Image-based Sequence Recognition and Its Application
- End-to-End Reinforcement Learning of Dialogue Agents for Information Access
- 论文阅读:Joint Learning of Single-image and Cross-image Representations for Person Re-identification
- 160 You want to use the automatic management of backup and recovery operations features for your dat
- 论文笔记:Improving Deep Visual Representation for Person Re-identification by Global and Local Image-language Association