首页 > 其他分享 >Checkerboard Context Model for Efficient Learned Image Compression

Checkerboard Context Model for Efficient Learned Image Compression

时间:2023-11-04 22:23:35浏览次数:38  
标签:上下文 Compression Efficient Image boldsymbol Context aligned hat 模型

目录

Abstract

自回归模型有效提高了RD表现,因为它有效减少了潜表示的空间冗余,但其解码时需要按照特定的顺序,而不能并行。本文的棋盘上下文模型,重新组织解码顺序,解码速度快了40倍。

Introduction

减少冗余的三个途径,空间、视觉和统计冗余。JPEG,JPEG2000 and BPG 都是用无损熵编码,内容损失只发生在量化阶段。
image
图1为空间因果上下文的掩码卷积模型,红色块代表要编码/解码的部分,Latents in yellowand blue locations are currently visible (all of them are visible during encoding, and those who have been decoded are visible during decoding).上下文模型可使用掩码卷积来进行上下文建模,以红色块为中心,与黄色块做卷积。
图(a)(b)是常用的串行上下文模型。(c)(d)是棋盘模型,当黄色与蓝色位置解码后,其他位置都可以并行解码。

Preliminary 初步介绍

image
尺度超先验框架如图2所示,
\(\quad\) \(g_s\),\(h_a\),\(h_s\)是神经网络实现的nonlinear transforms
\(\quad\) \(x\)是原始图像
\(\quad\) \(y=g_a(x)\)和\(\hat{y}=Q(y)\),分别是量化前后的 latent representations
\(\quad\) \(\hat{x}=g_s(\hat{y})\)是重构图像。
\(\quad\) \(z=h_a(y)\)和\(\hat{z}=Q(z)\),分别是量化前后的hyper latent
\(\quad\) \(\hat{z}\)被用作边信息,为latent \(\hat{y}\)的熵模型估计尺度参数\(\sigma\)
在训练时,通过加均匀噪声来近似量化操作,得到可微的\(\tilde{y},\tilde{z},\tilde{x}\)
The tradeoff between rate and distortion(loss function):

\[\begin{aligned} \begin{aligned}R+\lambda\cdot D\end{aligned}& \begin{aligned}&=\mathbb{E}_{\boldsymbol{x}\sim p_{\boldsymbol{x}}}[-\log_{2}p_{\boldsymbol{\hat{y}}|\boldsymbol{\hat{z}}}(\boldsymbol{\hat{y}}|\boldsymbol{\hat{z}})-\log_{2}p_{\boldsymbol{\hat{z}}}(\boldsymbol{\hat{z}})]\end{aligned} \\ &+\lambda\cdot\mathbb{E}_{\boldsymbol{x}\sim p_{\boldsymbol{x}}}|d(\boldsymbol{x},\boldsymbol{x})| \end{aligned}\]

其中,\(\hat{y}\) 和 hyper latent \(\hat{z}\)通过估计熵来预测, \(\lambda\) controls the bit rate,\(\lambda\)越大,图像重构质量越好。D通常用评价指标MSE或者MS-SSIM.E是指数分布
使用超先验尺度scale hyperprior,latents, \(\hat{y}\)的概率可以通过条件高斯尺度混合模型(GSM)来进行建模:

\[\begin{aligned}p_{\boldsymbol{\hat{y}}|\boldsymbol{\hat{z}}}(\hat{y}_i|\boldsymbol{\hat{z}})&=\left[\mathcal{N}(\mu_i,\sigma_i^2)*\mathcal{U}(-\frac12,\frac12)\right](\hat{y}_i)\end{aligned} \]

\[p_{\boldsymbol{\hat{y}}|\boldsymbol{\hat{z}}}(\boldsymbol{\hat{y}}|\boldsymbol{\hat{z}})=\prod_ip_{\boldsymbol{\hat{y}}|\boldsymbol{\hat{z}}}(\hat{y}_i|\boldsymbol{\hat{z}}) \]

其中,位置参数\(\mu_i\)假设为0,尺度参数\(\sigma_i\)是\(\sigma=h_s(\hat{z})\)第i个元素,因为\(\hat{y}\)中每个代码给定了超先验。hyper latent \(\hat{z}\)的概率可以用为参数全因分解密度模型建模。

Variational Image Compression with Hyperprior(超先验变分图像压缩)

Autoregressive Context(自回归上下文模型)

Parallel Context Modeling 并行上下文模型

Random-Mask Model: Test Arbitrary Masks(随机掩码模型)

How Distance Influences Rate Saving

Parallel Decoding with Checkerboard Context(棋盘模型并行解码)

标签:上下文,Compression,Efficient,Image,boldsymbol,Context,aligned,hat,模型
From: https://www.cnblogs.com/CLGYPYJ/p/17809363.html

相关文章

  • [论文阅读] Painterly Image Harmonization using Diffusion Model
    Pretitle:PainterlyImageHarmonizationusingDiffusionModelaccepted:AAAI2023paper:https://arxiv.org/abs/2212.08846code:https://github.com/bcmi/PHDiffusion-Painterly-Image-Harmonizationref:https://mp.weixin.qq.com/s/0AzaD8qVOFJrFeeIaJ4sTgref:h......
  • 图片预览功能实现,放大缩小(el-image-viewer)
    <template><div><divv-for="(item,index)inlist":key="index"class="box"><img:src="item"alt=""@click="onPreview(item)"/></div&......
  • java使用ImageIO读取CMYK图片转存为RGB图片在本地和线上表现不同的问题
    项目里有jpg图片是CMYK颜色模式,需要转成RGB颜色模式,我使用的方法简单粗暴,就是利用ImageIO转存一下,在我的本地正常。但是丢到服务器上就有问题了,色差很大。不知道是什么情况。`BufferedImageimg=ImageIO.read(src);ImageIO.write(img,"jpeg",dest);`本......
  • 【Azure K8S | AKS】在中国区AKS上遇见ImagePullBackOff时的替代方案
    Failedtopullimage"k8s.gcr.io/cluster-proportional-autoscaler-amd64:1.1.2-r2":rpcerror:code=Unknowndesc=Errorresponsefromdaemon:Gethttps://k8s.gcr.io/v2/:net/http:requestcanceledwhilewaitingforconnection(......
  • ELIC: Efficient Learned Image Compression with Unevenly Grouped Space-Channel Co
    abstruct\(\quad\)受能量压缩表现的启发,提出了不均匀通道情况自适应编码.结合不均匀分组模型和现有上下文模型,获得一种空间通道上下文自适应模型,来提高编码性能,而不影响其运行时间。\(\quad\)这种模型支持预览解码和渐进解码。introduction学习图像压缩中最重要的技术联合前......
  • 【CVPR2023】Efficient and Explicit Modelling of Image Hierarchies for Image Rest
    >论文:https://readpaper.com/paper/4728855966703960065代码:https://github.com/ofsoundof/GRL-Image-Restoration这个论文的代码地址叫GRL,意思是Global,Regional,Local的意思,作者从三个尺度对特征建模,核心是构建了一个anchoredstripself-attention。如何从Global,R......
  • 【CVPR2023】Learning A Sparse Transformer Network for Effective Image Deraining
    论文:https://readpaper.com/paper/4736105248993591297代码:https://github.com/cschenxiang/DRSformerTransformer模型通常使用标准的QKV三件套进行计算,但是部分来自K的token与来自Q的token并不相关,如果仍然对这些token进行特征聚合计算会影响图像修复的性能。......
  • dom-to-image图像失真
    //height:document.getElementById("hwLabelform").scrollHeight,//canvas高//width:document.getElementById("hwLabelform").scrollWidth,//canvas宽//scale:4,//按比例增加分辨率(2=双倍).domtoimage.toPng(document.getElementById('table'......
  • MITK编译错误C2220 mitkLabelSetImageToSurfaceFilter.cpp
    错误 C2220 以下警告被视为错误(编译源文件E:\0_MITK\MITK\Modules\Multilabel\mitkLabelSetImageToSurfaceFilter.cpp)[E:\0_MITK\MITK\SuperBuild\MITK-build\Modules\Multilabel\MitkMultilabel.vcxproj] MITK-build E:\0_MITK\MITK\SuperBuild\ep\include\ITK-5.2\i......
  • lesson16-ImageIconDemo
     在Eclipse里面执行   packageZYM_GUI;importjava.awt.Container;importjava.net.URL;importjavax.swing.*;publicclassImageIconDemoextendsJFrame{ publicImageIconDemo(){ JLabeljLabel=newJLabel("Image"); ......