最新:
重要:
PPT:https://blog.csdn.net/f290131665/article/details/79572012
评价指标:https://blog.csdn.net/f290131665/article/details/79514410
AOD-NET:
https://www.pytorchtutorial.com/pytorch-image-dehazing/
Benchmarking Single Image Dehazing and Beyond
去除雨滴,去雾,去除噪声,去尘土和去模糊等都是这一类的,图像复原(低级图像处理/视觉任务)。 采用生成对抗网络和感知损失进行这类研究,也已经很多很多。 以下是一些工作,但是未必采用GAN去做。 1、Density-aware Single Image De-raining using a Multi-stream Dense Network CVPR2018 有感知/特征损失,[paper]、[testing code] 密度感知多路密集网络DID-MDN,联合完成雨点密度估计和雨点去除。 图 1首先对输入图像的雨点程度(严重、中等、轻微)进行分类/估计,然后利用多路密集网络和标注信息对输入图像进行去雨处理。 效果非常好,速度也是非常快,应该是目前最好的模型。算法中的预训练,然后联合训练,估计是很tricky的。 2、Attentive Generative Adversarial Network for Raindrop Removal from a Single Image CVPR2018 [paper] 图 2该模型基于pix2pix,增加了attention-recurrent network,效果上比eigen2013的论文(第一个使用DL解决该问题的工作)好,也比pix2pix好。但是给论文没有和其他算法比。 3、Densely Connected Pyramid Dehazing Network CVPR2018 去雾,有感知损失,[paper]、[code] 图 3使用黄色网络估计transmission,利用蓝色网络估计atmospheric light,然后利用公司,计算得到去雾图像。论文中总损失有4个子损失,训练非常tricky。。。 4、Deep Joint Rain Detection and Removal from a Single Image CVPR2017 [paper]、[code] 比1差。 5、Image De-raining Using a Conditional Generative Adversarial Network 2017 类似pix2pix,[paper]、[code] 1中作者的以前工作,类似pix2pix。 6、Clearing the Skies: A Deep Network Architecture for Single-Image Rain Removal TIP2017 [paper] 7、Removing rain from single images via a deep detail network CVPR2017 [paper] 8、Rain Streak Removal Using Layer Priors CVPR2016 [paper] Single Image Rain Streak Decomposition Using Layer Priors TIP2017 [paper] 9、Perceptual Adversarial Networks for Image-to-Image Transformation 2017 类似pix2pix,有感知损失,[paper] |
- 顶级图像去雾算法效果讨论:http://bbs.csdn.net/topics/390573504
- 《Single Image Haze Removal Using Dark Channel Prior》一文中图像去雾算法的原理、实现、效果及其他
- Single Image Haze Removal(图像去雾)
- 论文翻译:https://blog.csdn.net/lee_cv/article/details/17280479
- 雾霾太重?深度神经网络教你如何图像去雾
- DehazeNet: An End-to-End System for Single Image Haze Removal
- 一种可实时处理 O(1)复杂度图像去雾算法的实现
- 暗通道优先的图像去雾算法(下)
- OPTIMIZED CONTRAST ENHANCEMENT FOR REAL-TIME IMAGE AND VIDEO DEHAZING:http://mcl.korea.ac.kr/projects/dehazing/
- 图像去雾算法(一)相关研究及链接
- https://zhuanlan.zhihu.com/p/28875405
- 图像去雾技术综述
- 一年去雾算法研究的总结
- 双边滤波与导向滤波
http://blog.csdn.net/pi9nc/article/details/26592377 - S.G. Narasimhan and S.K. Nayar, 多幅图像(同一场景不同时间、天气)去雾 主页
- NASA, Retinex理论增强,主页。
Ana Belén Petro总结了NASA的Retinex理论,源代码,不过不是matlab版本的。 - Kopf,Deep Photo: Model-Based Photograph Enhancement and Viewing,3D场景去雾,没有源码。主页地址
- Fattal, single image dehazing, 主页*matlab代码*
- Fattal 2014,Automatic Recovery of the Atmospheric Light in Hazy Images,大气光恢复去雾,有代码,主页
- Fattal 2014,Dehazing using Color-Lines,无代码,主页
这里有个Matlab script converting jet-color images into [0,1] transmission values 主页 - Tarel,Fast visibility restoration from a single color or gray level image,matlab代码*实验主页*
-
He kaiming, single image dehazing using dark channel prior,实验主页
其guided image dehazing,主页还有matlab代码 -
Nishino,bayesian defogging,贝叶斯去雾,主页
- Ancuti,inverse-image dehazing, fusion-based dehazing,水下融合去雾,个人主页*半反去雾主页*
- Ketan Tang, 基于学习的去雾Investigating haze-relevant features in a learning framework for image dehazing, 实验主页
- Gibson,维纳滤波去雾,fast single image fog removal using the adaptive wiener filter,主页
- Meng gaofeng,改进的暗原色去雾efficinet image dehazing with boundary constraint contextual regularization,matlab代码
- Yoav Y.Schechner,一直研究偏振去雾算法,典型的代表作,blind haze separation, advanced visiblity improvement based on polarization filtered images,主页
- yk wang,Single Image Defogging by Multiscale Depth Fusion,也是基于贝叶斯和马尔可夫来去雾,暂时没公布matlab代码。主页
-
Jin-Hwan Kim, optimized contrast enhancement for real-time image and video dehazing, 关于图像增强和视频去雾的,主页有代码,但是是C程序的。主页
-
ECCV2016 Single Image Dehazing via Multi-Scale Convolutional Neural Networks,主页
-
2015年的一篇CVPR,主页有代码,Simultaneous Video Defogging and Stereo Reconstruction 链接
-
此外还有文章主页:http://www.lizhuwen.com/pages/Stereo%20in%20Fog.html
关于去雾算法质量评价对比
1、Zhengying Chen,Quality Assessment for Comparing Image Enhancement Algorithms(CVPR2014),基于学习的去雾算法排序方法,据说有数据库,但得填表找他们要,主页
2、Gibson,A No-Reference Perceptual Based Contrast Enhancement Metric for Ocean Scenes in Fog(TIP,2013),一种CEM评价方法,不过也是基于学习的,数据库和代码都有。主页
3、Hautiere,Blind contrast enhancement assessment by gradient ratioing at visible edges,三种忙评价方法。代码网络上有,原作者编写的在这里,主页
4.图像去雾和视频去雾的综述,感觉整理的还蛮全的,Review of Video and Image Defogging Algorithms and Related Studies on Image Restoration and Enhancement。另外整理了部分代码。
python实现:
https://blog.csdn.net/chongshangyunxiao321/article/details/50997111
https://blog.csdn.net/zmshy2128/article/details/53443227
https://www.cnblogs.com/x1mercy/p/7824678.html
我需要的:
https://blog.csdn.net/zmshy2128/article/details/53443227 #python实现 :暗通道去雾算法的python实现
2008_CVPR_Tan_Visibility in Bad Weather from A Single Image
2008_Siggraph_Single Image Dehazing
2008_Siggraph Asia_Deep Photo: Model-Based Photograph Enhancement and Viewing
2009_CVPR_Single Image Haze Removal using Dark Channel Prior
2009_ICCV_Fast Visibility Restoration from a Single Color or Gray Level Image
2009_ICCV_Factorizing Scene Albedo and Depth from a Single Foggy Image
2012_ICPR_An Edge-preserving Filtering Framework for Visibility Restoration
2013_Optimized contrast enhancement for real-time image and video dehazing
2013_ICCV_Efficient Image Dehazing with Boundary Constraint and Contextual Regularization
2013_ICCV_Polarization-based dehazing using two reference objects
2014_CVPR_Investigating Haze-relevant Features in A Learning Framework for Image Dehazing
《Single Image Haze Removal Using Dark Channel Prior》一文中图像去雾算法的原理、实现、效果及其他。
香港城市大学-杨庆雄-Real-time O(1) Bilateral Filtering
Fast image defogging technology
数据集
1. D-hazy数据集
介绍了如何用深度图生成雾天
使用了以下室内数据集
middlebury
http://vision.middlebury.edu/stereo/data/scenes2014/
2.NYU2 Depth
https://cs.nyu.edu/~silberman/datasets/nyu_depth_v2.html
百度云链接及提取说明:https://blog.csdn.net/sinat_26871259/article/details/82351276
介绍 https://www.jianshu.com/p/870c541337b4
3. RESIDE
https://sites.google.com/view/reside-dehaze-datasets
1).训练集 ITS(indoor training set)
室内,合成
利用1399张middlebury 和NYU2 Depth室内深度数据集 生成13990个图。一个 真实值对应10个雾天图
分成13000训练集和990验证集
2). 测试集1 SOTS(synthetic objective testing set)
室内图 算法客观评价
从NYU2中选了500张(与训练集无重复),生成方式与训练集同。
2). 测试集2 HSTS(Hybrid Subjecive testing set)
室外图 人主观评价
10张合成图,与test1同。10张真实图。
4. RESIDE-beta(增加室外)
OTS(outdoor training set)
用2061张来自北京实时天气的真实室外图,使用【38】“Learning depth from single monocular images using deep convolutional neural fields,”中的算法减少深度的误差和可能生成的视觉造假现象。
beta属于【0.04,0.06,0.08,0.1,0.12,0.16,0.2】7类
A属于【0.8,0.85,0.9,0.95,1】5类
所以共生成572061=72135张
real world task-driven testing set
这里面有广汽研究院提供真实雾霾测试数据(CHINAMM比赛数据)
https://pan.baidu.com/s/1nuJOdjr 密码: n3v8