基于深度学习的图像修复论文合集
2020论文合集:https://mp.weixin.qq.com/s?__biz=MzIwMTE1NjQxMQ==&mid=2247519592&idx=2&sn=3a0598c9f52e47929678a572ea451d98&chksm=96f0ff3ca187762a107b4b9194e862b757d3d943ec399b35cbb7576cd92ee55cc648d7121ac3&scene=21#wechat_redirect
1.3D Photography Using Context-Aware Layered Depth Inpainting
2.Contextual Residual Aggregation for Ultra High-Resolution Image Inpainting
3.Bringing Old Photos Back to Life
4.Assessing Eye Aesthetics for Automatic Multi-Reference Eye In-Painting
其他论文:
1. CVPR 2016的Context-Encoders(CNN+GAN, 鼻祖级的 NN修复方法)
链接: Feature Learning by Inpainting;
Github代码:
pathak22/context-encodergithub.com
2. CVPR 2017的High Resolution Inpainting(Context-Encoders+CNNMRF)
链接: High-Resolution Image Inpainting using Multi-Scale Neural Patch Synthesis;
Github代码:
leehomyc/Faster-High-Res-Neural-Inpaintinggithub.com
3. ICCV 2017的on demanding learning(感觉也是Context-Encoders的衍生版...)
链接:
On-Demand Learning for Deep Image Restoration,
Github代码:
rhgao/on-demand-learninggithub.com
4. SIGGRAPH 2017 (ACM ToG)的Globally and Locally Consistent Image Completion
(CE中加入Global+Local两个判别器的改进),
Github代码:
1)https://github.com/satoshiiizuka/siggraph2017_inpaintinggithub.com
2)https://github.com/shinseung428/GlobalLocalImageCompletion_TF
5. ICLR 2018的New AI Imaging Technique Reconstructs Photos with Realistic Results
Image Inpainting for Irregular Holes UsingPartial Convolutions
号称秒杀PS的AI图像修复神器,来自于Nvidia 研究团队。引入了局部卷积,能够修复任意非中心、不规则区域),代码还没有放出来
[1804.07723] Image Inpainting for Irregular Holes Using Partial Convolutionsarxiv.org
6. CVPR 2018的Generative Image Inpainting with Contextual Attention,
一作大佬jiahui Yu 后续还有个工作: Free-Form Image Inpainting with Gated Convolution,
Github代码:
JiahuiYu/generative_inpaintinggithub.com
7. 哈工大左旺孟老师他们也有一篇Shift-Net: Image Inpainting via Deep Feature Rearrangement
8.Deep image prior
项目主页:https://dmitryulyanov.github.io/deep_image_prior
适用场景:
1)难以建模图像退化过程
2)难以得到训练图像进行监督训练
ECCV 2018的Contextual-based Image Inpainting,inpainting大佬Chao Yang(NPS的一作)等人的又一力作:
Contextual-based Image Inpaintingarxiv.org
10.ArXiv 2019 EdgeConnect:使用对抗边缘学习进行生成图像修复
项目地址:https://github.com/knazeri/edge-connect#testing
11. ACM MM 2018的Semantic Image Inpainting with Progressive Generative Networks,简称PGN,采用了由外至内的步进式修补策略,Github代码:
crashmoon/Progressive-Generative-Networksgithub.com
12. NIPS 2018的Image Inpainting via Generative Multi-column Convolutional Neural Networks,用了不少trick,
Github代码:
shepnerd/inpainting_gmcnngithub.com
13. CVPR 2019的Foreground-aware Image Inpainting, 思路类似于上面的工作,也是先推断生成轮廓边缘,辅助缺失区域进行修复,不知道上面的哥们看了这篇会是什么感受...速度也很重要啊...
Foreground-aware Image Inpaintingarxiv.org
14. CVPR 2019的Pluralistic Image Completion,
论文与Github代码:https://arxiv.org/abs/1903.04227arxiv.org
lyndonzheng/Pluralistic-Inpaintinggithub.com
15. IJCAI 2019的MUSICAL: Multi-Scale Image Contextual Attention Learning for Inpainting,武汉大学杜博老师组的工作(注:第一作者为我校计院的一名本科生...广大CV狗瑟瑟发抖!)。引入一个多尺度的上下文注意力模块,避免信息滥用/误用导致的纹理模糊等问题,损失函数部分联合了风格损失、感知损失、对抗损失,来保证补绘内容的一致性和清晰水平。
武汉大学地学智能感知与机器学习研究组sigma.whu.edu.cn
16. ArXiv 2019的Coherent Semantic Attention for Image Inpainting,论文作者为@Kuma , 文中提出了一个全新的Attention模块,该模块不仅有效的利用了上下文信息同时能够捕捉到生成补丁之间的相关性。同时提出了一个新的损失函数配合模块的工作,最后利用一个新的特征感知辨别器对细节效果进行加强,代码过段时间会公开。
Coherent Semantic Attention for Image Inpaintingarxiv.org
KumapowerLIU - Overviewgithub.com

浙公网安备 33010602011771号