No Cover Image

Journal article 199 views 1 download

A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation

Tianyu Zhang Orcid Logo, Fan Wan, Xingyu Miao Orcid Logo, Jingjing Deng Orcid Logo, Xianghua Xie Orcid Logo, Yang Long Orcid Logo

Pattern Recognition, Volume: 180, Issue: Part B, Start page: 113915

Swansea University Author: Xianghua Xie Orcid Logo

  • 71874.VOR.pdf

    PDF | Version of Record

    This is an open access article distributed under the terms of the Creative Commons CC-BY license.

    Download (2.6MB)

Check full text

DOI (Published version): 10.1016/j.patcog.2026.113915

Abstract

Dynamic convolution is an advanced deep-learning strategy that enables neural networks to adjust their convolutional kernels dynamically in response to varying input data. This adaptability enhances the network’s efficiency in processing diverse features. However, traditional dynamic convolution tec...

Full description

Published in: Pattern Recognition
ISSN: 0031-3203 1873-5142
Published: Elsevier BV 2026
Online Access: Check full text

URI: https://cronfa.swan.ac.uk/Record/cronfa71874
first_indexed 2026-05-11T10:08:06Z
last_indexed 2026-06-20T05:02:21Z
id cronfa71874
recordtype SURis
fullrecord <?xml version="1.0"?><rfc1807><datestamp>2026-06-19T14:53:41.4255343</datestamp><bib-version>v2</bib-version><id>71874</id><entry>2026-05-11</entry><title>A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation</title><swanseaauthors><author><sid>b334d40963c7a2f435f06d2c26c74e11</sid><ORCID>0000-0002-2701-8660</ORCID><firstname>Xianghua</firstname><surname>Xie</surname><name>Xianghua Xie</name><active>true</active><ethesisStudent>false</ethesisStudent></author></swanseaauthors><date>2026-05-11</date><deptcode>MACS</deptcode><abstract>Dynamic convolution is an advanced deep-learning strategy that enables neural networks to adjust their convolutional kernels dynamically in response to varying input data. This adaptability enhances the network&#x2019;s efficiency in processing diverse features. However, traditional dynamic convolution techniques often overlook the critical role of local features in image classification, resulting in suboptimal performance in capturing fine details and textures necessary for accurate image analysis. To address this, our research introduces Adaptive Attention-Driven Dynamic Convolution (A2D2C), an innovative adaptive adjustment mechanism that focuses on local image features, significantly improving the network&#x2019;s ability to capture fine details and overall performance. Moreover, our paper proposes a novel dynamic convolution that enhances the network&#x2019;s feature learning ability by combining the input feature map with multiple convolution kernels to generate the attention weights. Additionally, we develop a streamlined version of our model, named A2D2C+, which significantly increases operational efficiency and reduces computational costs. Experimental evaluations on the ImageNet, CIFAR-100 and COCO datasets demonstrate substantial performance enhancements, underscoring the efficacy and applicability of our approach.</abstract><type>Journal Article</type><journal>Pattern Recognition</journal><volume>180</volume><journalNumber>Part B</journalNumber><paginationStart>113915</paginationStart><paginationEnd/><publisher>Elsevier BV</publisher><placeOfPublication/><isbnPrint/><isbnElectronic/><issnPrint>0031-3203</issnPrint><issnElectronic>1873-5142</issnElectronic><keywords>Attention; Dynamic convolution; Local features</keywords><publishedDay>1</publishedDay><publishedMonth>12</publishedMonth><publishedYear>2026</publishedYear><publishedDate>2026-12-01</publishedDate><doi>10.1016/j.patcog.2026.113915</doi><url/><notes/><college>COLLEGE NANME</college><department>Mathematics and Computer Science School</department><CollegeCode>COLLEGE CODE</CollegeCode><DepartmentCode>MACS</DepartmentCode><institution>Swansea University</institution><apcterm>Another institution paid the OA fee</apcterm><funders>This work was supported by the Royal Society International Exchanges Scheme&#x2014;Towards Collaborative Cloud-Edge Deep Learning Deployment under Grant IEC/NSFC/223523; the National Edge AI Hub for Real Data: Edge Intelligence for Cyber-disturbances and Data Quality under Grant EP/Y028813/1; and the UK Medical Research Council (MRC) Innovation Fellowship under Grant MR/S003916/2.</funders><projectreference/><lastEdited>2026-06-19T14:53:41.4255343</lastEdited><Created>2026-05-11T10:59:39.5273224</Created><path><level id="1">Faculty of Science and Engineering</level><level id="2">School of Mathematics and Computer Science - Computer Science</level></path><authors><author><firstname>Tianyu</firstname><surname>Zhang</surname><orcid>0009-0007-2866-8386</orcid><order>1</order></author><author><firstname>Fan</firstname><surname>Wan</surname><order>2</order></author><author><firstname>Xingyu</firstname><surname>Miao</surname><orcid>0000-0003-1203-8279</orcid><order>3</order></author><author><firstname>Jingjing</firstname><surname>Deng</surname><orcid>0000-0001-9274-651x</orcid><order>4</order></author><author><firstname>Xianghua</firstname><surname>Xie</surname><orcid>0000-0002-2701-8660</orcid><order>5</order></author><author><firstname>Yang</firstname><surname>Long</surname><orcid>0000-0002-2445-6112</orcid><order>6</order></author></authors><documents><document><filename>71874__37018__293ac426380146b4b30ac60e6ada8289.pdf</filename><originalFilename>71874.VOR.pdf</originalFilename><uploaded>2026-06-19T14:51:14.0322690</uploaded><type>Output</type><contentLength>2726762</contentLength><contentType>application/pdf</contentType><version>Version of Record</version><cronfaStatus>true</cronfaStatus><documentNotes>This is an open access article distributed under the terms of the Creative Commons CC-BY license.</documentNotes><copyrightCorrect>true</copyrightCorrect><language>eng</language><licence>http://creativecommons.org/licenses/by/4.0/</licence></document></documents><OutputDurs/></rfc1807>
spelling 2026-06-19T14:53:41.4255343 v2 71874 2026-05-11 A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation b334d40963c7a2f435f06d2c26c74e11 0000-0002-2701-8660 Xianghua Xie Xianghua Xie true false 2026-05-11 MACS Dynamic convolution is an advanced deep-learning strategy that enables neural networks to adjust their convolutional kernels dynamically in response to varying input data. This adaptability enhances the network’s efficiency in processing diverse features. However, traditional dynamic convolution techniques often overlook the critical role of local features in image classification, resulting in suboptimal performance in capturing fine details and textures necessary for accurate image analysis. To address this, our research introduces Adaptive Attention-Driven Dynamic Convolution (A2D2C), an innovative adaptive adjustment mechanism that focuses on local image features, significantly improving the network’s ability to capture fine details and overall performance. Moreover, our paper proposes a novel dynamic convolution that enhances the network’s feature learning ability by combining the input feature map with multiple convolution kernels to generate the attention weights. Additionally, we develop a streamlined version of our model, named A2D2C+, which significantly increases operational efficiency and reduces computational costs. Experimental evaluations on the ImageNet, CIFAR-100 and COCO datasets demonstrate substantial performance enhancements, underscoring the efficacy and applicability of our approach. Journal Article Pattern Recognition 180 Part B 113915 Elsevier BV 0031-3203 1873-5142 Attention; Dynamic convolution; Local features 1 12 2026 2026-12-01 10.1016/j.patcog.2026.113915 COLLEGE NANME Mathematics and Computer Science School COLLEGE CODE MACS Swansea University Another institution paid the OA fee This work was supported by the Royal Society International Exchanges Scheme—Towards Collaborative Cloud-Edge Deep Learning Deployment under Grant IEC/NSFC/223523; the National Edge AI Hub for Real Data: Edge Intelligence for Cyber-disturbances and Data Quality under Grant EP/Y028813/1; and the UK Medical Research Council (MRC) Innovation Fellowship under Grant MR/S003916/2. 2026-06-19T14:53:41.4255343 2026-05-11T10:59:39.5273224 Faculty of Science and Engineering School of Mathematics and Computer Science - Computer Science Tianyu Zhang 0009-0007-2866-8386 1 Fan Wan 2 Xingyu Miao 0000-0003-1203-8279 3 Jingjing Deng 0000-0001-9274-651x 4 Xianghua Xie 0000-0002-2701-8660 5 Yang Long 0000-0002-2445-6112 6 71874__37018__293ac426380146b4b30ac60e6ada8289.pdf 71874.VOR.pdf 2026-06-19T14:51:14.0322690 Output 2726762 application/pdf Version of Record true This is an open access article distributed under the terms of the Creative Commons CC-BY license. true eng http://creativecommons.org/licenses/by/4.0/
title A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation
spellingShingle A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation
Xianghua Xie
title_short A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation
title_full A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation
title_fullStr A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation
title_full_unstemmed A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation
title_sort A2D2C: Adaptive attention-driven dynamic convolution for local feature adaptation
author_id_str_mv b334d40963c7a2f435f06d2c26c74e11
author_id_fullname_str_mv b334d40963c7a2f435f06d2c26c74e11_***_Xianghua Xie
author Xianghua Xie
author2 Tianyu Zhang
Fan Wan
Xingyu Miao
Jingjing Deng
Xianghua Xie
Yang Long
format Journal article
container_title Pattern Recognition
container_volume 180
container_issue Part B
container_start_page 113915
publishDate 2026
institution Swansea University
issn 0031-3203
1873-5142
doi_str_mv 10.1016/j.patcog.2026.113915
publisher Elsevier BV
college_str Faculty of Science and Engineering
hierarchytype
hierarchy_top_id facultyofscienceandengineering
hierarchy_top_title Faculty of Science and Engineering
hierarchy_parent_id facultyofscienceandengineering
hierarchy_parent_title Faculty of Science and Engineering
department_str School of Mathematics and Computer Science - Computer Science{{{_:::_}}}Faculty of Science and Engineering{{{_:::_}}}School of Mathematics and Computer Science - Computer Science
document_store_str 1
active_str 0
description Dynamic convolution is an advanced deep-learning strategy that enables neural networks to adjust their convolutional kernels dynamically in response to varying input data. This adaptability enhances the network’s efficiency in processing diverse features. However, traditional dynamic convolution techniques often overlook the critical role of local features in image classification, resulting in suboptimal performance in capturing fine details and textures necessary for accurate image analysis. To address this, our research introduces Adaptive Attention-Driven Dynamic Convolution (A2D2C), an innovative adaptive adjustment mechanism that focuses on local image features, significantly improving the network’s ability to capture fine details and overall performance. Moreover, our paper proposes a novel dynamic convolution that enhances the network’s feature learning ability by combining the input feature map with multiple convolution kernels to generate the attention weights. Additionally, we develop a streamlined version of our model, named A2D2C+, which significantly increases operational efficiency and reduces computational costs. Experimental evaluations on the ImageNet, CIFAR-100 and COCO datasets demonstrate substantial performance enhancements, underscoring the efficacy and applicability of our approach.
published_date 2026-12-01T06:02:21Z
_version_ 1868490857327362048
score 11.109323

Similar Items