Image Inpainting Based on High-Order Spatial Interaction Modules

Jianwu Li, Qihang Hu, Sichun Qin

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Image inpainting is a fundamental task in the field of computer vision, aimed at using partially visible pixels in an image to repair missing or damaged parts of the image. In this paper, we propose a multi-stage cascaded image inpainting model based on high-order spatial interaction modules, which can achieve input adaption and long-range spatial interaction by introducing recursive gated convolutions in each high-order spatial interaction module, resulting in a global perceptual field and effective modeling of correlation between the inpainted region and the visible image range. Extensive experiments on CelebA-HQ dataset and Paris StreetView dataset demonstrate the effectiveness of the proposed method.

Original languageEnglish
Title of host publicationICNC-FSKD 2024 - 20th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery
EditorsMaozhen Li, Ning Xiong, Jianguo Chen, Zheng Xiao, Kenli Li, Lipo Wang
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9798350356328
DOIs
Publication statusPublished - 2024
Event20th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery, ICNC-FSKD 2024 - Guangzhou, China
Duration: 27 Jul 202429 Jul 2024

Publication series

NameICNC-FSKD 2024 - 20th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery

Conference

Conference20th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery, ICNC-FSKD 2024
Country/TerritoryChina
CityGuangzhou
Period27/07/2429/07/24

Fingerprint

Dive into the research topics of 'Image Inpainting Based on High-Order Spatial Interaction Modules'. Together they form a unique fingerprint.

Cite this