当前位置:网站首页>对“Image Denoising Using an Improved Generative Adversarial Network with Wasserstein Distance“的理解
对“Image Denoising Using an Improved Generative Adversarial Network with Wasserstein Distance“的理解
2022-07-28 12:49:00 【RrS_G】
译:"基于Wasserstein距离的改进生成对抗网络图像去噪"
-- Proceedings of the 40th Chinese Control Conference -- 2021
一、概括
这篇文章提出了一种基于WGAN-GP的图像去噪算法。该方法可以从噪声图像中直接生成干净的图像,并能很好地保留图像的纹理细节。利用WGAN-GP损失来对抗损失,避免训练过程中梯度消失,使训练更加稳定,并在生成网络中加入感知损失来提高生成的图像质量,降低计算成本。
二、方法
2.1、降噪模型
首先,该模型的目的是找到一个G,将退化图像z映射成干净图像x:
![]()
2.2、损失函数
损失如下:

其中:

2.3、网络结构
结构如下:

三、实验结果

边栏推荐
- Excellent performance! Oxford, Shanghai, AI Lab, Hong Kong University, Shangtang, and Tsinghua have joined forces to propose a language aware visual transformer for reference image segmentation! Open
- R语言ggplot2可视化:可视化散点图并为散点图中的数据点添加文本标签、使用ggrepel包的geom_text_repel函数避免数据点标签互相重叠(自定义指定字体类型font family)
- R语言ggplot2可视化:使用ggpubr包的ggviolin函数可视化小提琴图、设置draw_quantiles参数添加指定分位数横线(例如,50%分位数、中位数)
- Cool operation preheating! Code to achieve small planet effect
- R language uses LM function to build linear regression model and subset function to specify subset of data set to build regression model (use floor function and length function to select the former pa
- [Architecture] reading notes of three micro service books with high scores
- .NET的求复杂类型集合的差集、交集、并集
- Rust from introduction to mastery 01 introduction
- Poj3275 ranking the cows
- C语言:顺序存储结构的快速排序
猜你喜欢

Excellent performance! Oxford, Shanghai, AI Lab, Hong Kong University, Shangtang, and Tsinghua have joined forces to propose a language aware visual transformer for reference image segmentation! Open

I'm bald! Who should I choose for unique index or general index?

持续(集成-->交付-->部署)

Customized template in wechat applet

酷炫操作预热!代码实现小星球特效

GO语言-栈的应用-表达式求值

Volcanic stone investment Zhang Suyang: hard technology, the relatively certain answer in the next 10 years

Cool operation preheating! Code to achieve small planet effect

org.apache.ibatis.exceptions.TooManyResultsException的异常排查过程

30 day question brushing plan (II)
随机推荐
Volcanic stone investment Zhang Suyang: hard technology, the relatively certain answer in the next 10 years
R language uses LM function to build multiple linear regression model, writes regression equation according to model coefficient, and uses conflict function to give 95% confidence interval of regressi
[C language] the difference between structure pointer and structure variable as formal parameters
Deployment之滚动更新策略。
Continuous (integration -- & gt; delivery -- & gt; deployment)
I'm bald! Who should I choose for unique index or general index?
要想组建敏捷团队,这些方法不可少
PHP generates random numbers (nickname random generator)
R language uses LM function to build linear regression model and subset function to specify subset of data set to build regression model (use floor function and length function to select the former pa
我秃了!唯一索引、普通索引我该选谁?
长封闭期私募产品再现 业内人士看法各异
Tutorial on the principle and application of database system (059) -- MySQL exercise questions: operation questions 1-10 (III)
111. SAP UI5 FileUploader 控件实现本地文件上传,接收服务器端的响应时遇到跨域访问错误
朋友发来几个面试题
JWT 登录认证 + Token 自动续期方案,写得太好了!
C语言:随机生成数+归并排序
R language ggplot2 visualization: use ggviolin function of ggpubr package to visualize violin diagram and set draw_ The quantiles parameter adds a specified quantile horizontal line (for example, 50%
基于神经网络的帧内预测和变换核选择
R语言因子数据的表格和列联表(交叉表)生成:使用summay函数分析列表查看卡方检验结果判断两个因子变量是否独立(使用卡方检验验证独立性)
Beyond Istio OSS——Istio服务网格的现状与未来