Deep Fusion Network for Image Completion

被引:67
作者
Hong, Xin [1 ]
Xiong, Pengfei [2 ]
Ji, Renhe [2 ]
Fan, Haoqiang [2 ]
机构
[1] Univ Chinese Acad Sci, Chinese Acad Sci, Inst Comp Technol, CAS Key Lab Network Data Sci & Technol, Beijing, Peoples R China
[2] Megvii Technol, Beijing, Peoples R China
来源
PROCEEDINGS OF THE 27TH ACM INTERNATIONAL CONFERENCE ON MULTIMEDIA (MM'19) | 2019年
关键词
image completion; inpainting; Deep Fusion Network; Fusion Block; alpha composition;
D O I
10.1145/3343031.3351002
中图分类号
TP39 [计算机的应用];
学科分类号
081203 ; 0835 ;
摘要
Deep image completion usually fails to harmonically blend the restored image into existing content, especially in the boundary area. This paper handles this problem from a new perspective of creating a smooth transition and proposes a concise Deep Fusion Network (DFNet). Firstly, a fusion block is introduced to generate a flexible alpha composition map for combining known and unknown regions. The fusion block not only provides a smooth fusion between restored and existing content but also provides an attention map to make network focus more on the unknown pixels. In this way, it builds a bridge for structural and texture information, so that information can be naturally propagated from the known region into completion. Furthermore, fusion blocks are embedded into several decoder layers of the network. Accompanied by the adjustable loss constraints on each layer, more accurate structure information is achieved. We qualitatively and quantitatively compare our method with other state-of-the-art methods on Places2 and CelebA datasets. The results show the superior performance of DFNet, especially in the aspects of harmonious texture transition, texture detail and semantic structural consistency.
引用
收藏
页码:2033 / 2042
页数:10
相关论文
共 38 条
[1]  
[Anonymous], EUR C COMP VIS
[2]  
[Anonymous], 2017, IEEE T PATTERN ANAL
[3]  
[Anonymous], 2016, EUR C COMP VIS
[4]  
[Anonymous], 2015, ARXIV PREPRINT ARXIV
[5]  
[Anonymous], 2018, P IEEE C COMP VIS PA
[6]  
[Anonymous], EUR C COMP VIS ECCV
[7]  
[Anonymous], 2018, IEEE C COMP VIS PATT
[8]  
[Anonymous], ACM MULTIMEDIA
[9]  
[Anonymous], 2016, COMPUTER VISION PATT
[10]  
[Anonymous], C WORKSH NEUR INF PR