Wenya Yang;Xiao-Diao Chen;Wen Wu;Hongshuai Qin;Kangming Yan;Xiaoyang Mao;Haichuan Song
{"title":"Boosting Deep Unsupervised Edge Detection via Segment Anything Model","authors":"Wenya Yang;Xiao-Diao Chen;Wen Wu;Hongshuai Qin;Kangming Yan;Xiaoyang Mao;Haichuan Song","doi":"10.1109/TII.2024.3376726","DOIUrl":null,"url":null,"abstract":"Segment anything model (SAM), a vision foundation network trained on a massive segmentation corpus, exhibits a superior boundary localization capability for nature images. This work aims to leverage such strengths to develop a deep unsupervised edge detection (UED) framework for alleviating the high reliance on dense labeling. However, applying vanilla SAM to edge detection fails to identify the salient edge cues but only the semantic boundary. This article introduces a lightweight adapter-tuning scheme to learn detailed edge information for filling the gap between boundary and edge, enabling a well-fitting even with limited training data. Moreover, considering the low-quality pseudo labels used in our UED framework, we propose two training strategies, adaptive progressive learning and gradient-guided pseudo label updating, to alleviate the impact of noisy labels from traditional UED methods. Extensive experiments demonstrate that our method achieves comparable results to state-of-the-art fully supervised edge detectors.","PeriodicalId":13301,"journal":{"name":"IEEE Transactions on Industrial Informatics","volume":null,"pages":null},"PeriodicalIF":11.7000,"publicationDate":"2024-04-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Transactions on Industrial Informatics","FirstCategoryId":"94","ListUrlMain":"https://ieeexplore.ieee.org/document/10490131/","RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"AUTOMATION & CONTROL SYSTEMS","Score":null,"Total":0}
引用次数: 0
Abstract
Segment anything model (SAM), a vision foundation network trained on a massive segmentation corpus, exhibits a superior boundary localization capability for nature images. This work aims to leverage such strengths to develop a deep unsupervised edge detection (UED) framework for alleviating the high reliance on dense labeling. However, applying vanilla SAM to edge detection fails to identify the salient edge cues but only the semantic boundary. This article introduces a lightweight adapter-tuning scheme to learn detailed edge information for filling the gap between boundary and edge, enabling a well-fitting even with limited training data. Moreover, considering the low-quality pseudo labels used in our UED framework, we propose two training strategies, adaptive progressive learning and gradient-guided pseudo label updating, to alleviate the impact of noisy labels from traditional UED methods. Extensive experiments demonstrate that our method achieves comparable results to state-of-the-art fully supervised edge detectors.
期刊介绍:
The IEEE Transactions on Industrial Informatics is a multidisciplinary journal dedicated to publishing technical papers that connect theory with practical applications of informatics in industrial settings. It focuses on the utilization of information in intelligent, distributed, and agile industrial automation and control systems. The scope includes topics such as knowledge-based and AI-enhanced automation, intelligent computer control systems, flexible and collaborative manufacturing, industrial informatics in software-defined vehicles and robotics, computer vision, industrial cyber-physical and industrial IoT systems, real-time and networked embedded systems, security in industrial processes, industrial communications, systems interoperability, and human-machine interaction.