图像修复(Image Inpainting)，顾名思义，就是将图像中损坏的部分修复起来。该技术可以应用在图像编辑上，例如移除物体(remove unwanted object), 图像补全，修复老照片等。传统的图像修复方法有diffusion-based和patch-based两种，而近些年来的方法多数都是基于深度学习来做的。今天就为大家整理2016-2022年图像修复领域的重要论文。
《Context Encoders: Feature Learning by Inpainting》——深度学习图像修复的开山之作
- 会议/期刊：CVPR 2016
- 论文链接：CVPR 2016 Open Access Repository
- 代码链接：GitHub - BoyuanJiang/context_encoder_pytorch: PyTorch Implement of Context Encoders: Feature Learning by Inpainting
- 作者：Deepak Pathak， Philipp Krahenbuhl， Jeff Donahue， Trevor Darrell， Alexei A. Efros
- 单位：University of California, Berkeley
《Globally and Locally Consistent Image Completion》——全局判别器和局部判别器进行训练
- 会议/期刊：ACM TOG 2017
- 论文链接：https://dl.acm.org/doi/abs/10.1145/3072959.3073659
- 代码链接：GitHub - otenim/GLCIC-PyTorch: A High-Quality PyTorch Implementation of “Globally and Locally Consistent Image Completion”.
- 作者：SATOSHI IIZUKA, EDGAR SIMO-SERRA, HIROSHI ISHIKAWA
- 单位：Waseda University
《Image Inpainting for Irregular Holes Using Partial Convolutions》——使用改进的部分卷积进行修复
- 会议/期刊：ECCV 2018
- 论文链接：ECCV 2018 Open Access Repository
- 代码链接：https://github.com/tanimutomo/partialconv
- 作者：Guilin Liu， Fitsum A. Reda， Kevin J. Shih， Ting-Chun Wang， Andrew Tao， Bryan Catanzaro
- 单位：NVIDIA Corporation
博文讲解: 卡卡猡特：详解Partial Convolution （一） | 图像修复领域经典之作 | 运算机制及模型结构
《Generative Image Inpainting with Contextual Attention》——注意力机制改进Contextual Attention
- 会议/期刊：CVPR 2018
- 论文链接：CVPR 2018 Open Access Repository
- 代码链接：https://github.com/JiahuiYu/generative_inpainting
- 作者：Jiahui Yu, Zhe Lin, Jimei Yang, Xiaohui Shen, Xin Lu, Thomas S. Huang
- 单位：University of Illinois at Urbana-Champaign, Adobe Research
《EdgeConnect: Structure Guided Image Inpainting using Edge Prediction》——根据结构边缘修复图像
- 会议/期刊：ICCV workshop 2019
- 论文链接：http://openaccess.thecvf.com/content_ICCVW_2019/html/AIM/Nazeri_EdgeConnect_Structure_Guided_Image_Inpainting_using_Edge_Prediction_ICCVW_2019_paper.html
- 代码链接：https://github.com/knazeri/edge-connect
- 作者：Kamyar Nazeri, Eric Ng, Tony Joseph, Faisal Z. Qureshi, and Mehran Ebrahimi
- 单位：University of Ontario Institute of Technology, Canada
《Free-Form Image Inpainting with Gated Convolution》——使用改进的Gated Convolution进行修复
- 会议/期刊：ICCV 2019
- 论文链接：https://openaccess.thecvf.com/content_ICCV_2019/html/Yu_Free-Form_Image_Inpainting_With_Gated_Convolution_ICCV_2019_paper.html
- 代码链接：https://github.com/JiahuiYu/generative_inpainting
- 作者：Jiahui Yu, Zhe Lin, Jimei Yang, Xiaohui Shen, Xin Lu, Thomas Huang
- 单位：University of Illinois at Urbana-Champaign, Adobe Research, ByteDance AI Lab
《Free-Form Image Inpainting with Gated Convolution》——使用改进的Gated Convolution进行修复
- 会议/期刊：ICCV 2019
- 论文链接：https://openaccess.thecvf.com/content_ICCV_2019/html/Yu_Free-Form_Image_Inpainting_With_Gated_Convolution_ICCV_2019_paper.html
- 代码链接：https://github.com/JiahuiYu/generative_inpainting
- 作者：Jiahui Yu, Zhe Lin, Jimei Yang, Xiaohui Shen, Xin Lu, Thomas Huang
- 单位：University of Illinois at Urbana-Champaign, Adobe Research, ByteDance AI Lab\
《Learning Pyramid-Context Encoder Network for High-Quality Image Inpainting》——金字塔式逐层修复
- 会议/期刊：CVPR 2019
- 论文链接：CVPR 2019 Open Access Repository
- 代码链接：https://github.com/researchmm/PEN-Net-for-Inpainting
- 作者：Yanhong Zeng, Jianlong Fu, Hongyang Chao, Baining Guo
- 单位：Sun Yat-sen University, Microsoft Research
《Pluralistic Image Completion》——提出图像多样化修复
- 会议/期刊：CVPR 2019
- 论文链接：CVPR 2019 Open Access Repository
- 代码链接：https://github.com/lyndonzheng/Pluralistic-Inpainting
- 作者：Chuanxia Zheng， Tat-Jen Cham， Jianfei Cai
- 单位：Nanyang Technological University, Singapore
《Bringing Old Photos Back to Life》——修复老照片
- 会议/期刊：CVPR 2020
- 论文链接：CVPR 2020 Open Access Repository
- 代码链接：GitHub - microsoft/Bringing-Old-Photos-Back-to-Life: Bringing Old Photo Back to Life (CVPR 2020 oral)
- 作者：Ziyu Wan1∗, Bo Zhang2, Dongdong Chen3, Pan Zhang4, Dong Chen2, Jing Liao1†, Fang Wen2
- 单位：1.City University of Hong Kong. 2.Microsoft Research Asia. 3.Microsoft Cloud + AI. 4.University of Science and Technology of China
《PD-GAN: Probabilistic Diverse GAN for Image Inpainting》——提出SPDNorm进行多样修复
- 会议/期刊：CVPR 2021
- 论文链接：CVPR 2021 Open Access Repository
- 代码链接：https://github.com/KumapowerLIU/PD-GAN
- 作者：Hongyu Liu1, Ziyu Wan2, Wei Huang3, Yibing Song4, Xintong Han1*, Jing Liao2
- 单位：1.Huya Inc 2.City University of Hong Kong 3.Hunan University 4.Tencent AI Lab
《High-Fidelity Pluralistic Image Completion with Transformers》——使用Transformer进行多样修复
- 会议/期刊：ICCV 2021
- 论文链接：ICCV 2021 Open Access Repository
- 代码链接：https://github.com/raywzy/ICT
- 作者：Ziyu Wan1, Jingbo Zhang1, Dongdong Chen2, Jing Liao1*
- 单位：1City University of Hong Kong 2Microsoft Cloud + AI
《CR-Fill: Generative Image Inpainting with Auxiliary Contextual Reconstruction》——提出语境重构辅助修复
- 会议/期刊：ICCV 2021
- 论文链接：https://openaccess.thecvf.com/content/ICCV2021/html/Zeng_CR-Fill_Generative_Image_Inpainting_With_Auxiliary_Contextual_Reconstruction_ICCV_2021_paper.html
- 代码链接：https://github.com/zengxianyu/crfill
- 作者：Yu Zeng1， Zhe Lin2， Huchuan Lu3,4， Vishal M. Patel1
- 单位：1. Johns Hopkins University 2. Adobe Research 3. Dalian University of Technology 4. Peng Cheng Lab
《Bridging Global Context Interactions for High-Fidelity Image Completion》——使用Transformer进行长距离注意力
- 会议/期刊：CVPR 2022
- 论文链接：Bridging Global Context Interactions for High-Fidelity Image Completion
- 代码链接：GitHub - lyndonzheng/TFill: CVPR2022:”Bridging Global Context Interactions for High-Fidelity Image Completion”
- 作者：Chuanxia Zheng, Tat-Jen Cham, Jianfei Cai, Dinh Phung
- 单位：Nanyang Technological University, Singapore, Monash University, Australia
《Image Inpainting with Local and Global Refinement》——通过全局及局部优化进行修复
- 会议/期刊：IEEE TIP 2022
- 论文链接：https://ieeexplore.ieee.org/abstract/document/9730792/
- 代码链接：https://github.com/weizequan/LGNet.git
- 作者：Weize Quan, Ruisong Zhang, Yong Zhang, Zhifeng Li, Jue Wang, and Dong-Ming Yan
- 单位：Institute of Automation, Chinese Academy of Sciences
《High-Quality Pluralistic Image Completion via Code Shared VQGAN》——使用VQGAN进行多样图像修复
- 会议/期刊：arxiv 2022
- 论文链接：https://arxiv.org/abs/2204.01931
- 代码链接：（待更新）
- 作者：Chuanxia Zheng,Guoxian Song,Tat-Jen Cham,Jianfei Cai,Dinh Phung,Linjie Luo
- 单位：Monash University, Australia；Nanyang Technological University, Singapore；ByteDance Inc, USA
《Aggregated Contextual Transformations for High-Resolution Image Inpainting》——提出了一种Aggregated COntextual-Transformation GAN的高分辨率图像修复方法
- 会议/期刊：IEEE TIP 2022
- 论文链接：https://ieeexplore.ieee.org/abstract/document/9729564
- 代码链接：https://github.com/researchmm/AOT-GAN-for-Inpainting
- 作者：Yanhong Zeng, Jianlong Fu, Hongyang Chao, and Baining Guo
- 单位：Microsoft Research, Sun Yat-sen University
