ReBiT-Net: Resource-Efficient Bidirectional Transmission Network for RGB-D Salient Object Detection

IF 1.6 4区 工程技术 Q3 ENGINEERING, ELECTRICAL & ELECTRONIC Journal of Electrical Engineering & Technology Pub Date : 2024-07-04 DOI:10.1007/s42835-024-01971-z
Youpeng Yi, Jiawei Xu, Xiaoqin Zhang, Seop Hyeong Park
{"title":"ReBiT-Net: Resource-Efficient Bidirectional Transmission Network for RGB-D Salient Object Detection","authors":"Youpeng Yi, Jiawei Xu, Xiaoqin Zhang, Seop Hyeong Park","doi":"10.1007/s42835-024-01971-z","DOIUrl":null,"url":null,"abstract":"<p>Existing artificial neural network-based methodologies for salient object detection in RGB-depth (RGB-D) images typically require significant memory and computation time. In this paper, we propose ReBiT-Net, an novel and resource-efficient network designed to addresses this issue. ReBiT-Net utilizes a mobile network for feature extraction and incorporates depth map quality to regulate the fusion of multi-modal features, resulting in top-to-bottom refinement of salient objects using salient information. Empirical evaluations conducted on five benchmark datasets demonstrate the superior performance of our model in terms of accuracy (achieving 334 frames per second for an input size of 320 <span>\\(\\times\\)</span> 320) and model parameters (merely 5.1 MB). Moreover, we introduce ReBiT-Net*, a simplified variant of ReBiT-Net, which entails reduced model parameters (4.2 MB) and enhanced processing speed (793 frames per second for a 256 <span>\\(\\times\\)</span> 256 input size). These improvements are achieved through reduced memory requirements and computational demands via the adoption of a smaller input image size.</p>","PeriodicalId":15577,"journal":{"name":"Journal of Electrical Engineering & Technology","volume":null,"pages":null},"PeriodicalIF":1.6000,"publicationDate":"2024-07-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Electrical Engineering & Technology","FirstCategoryId":"5","ListUrlMain":"https://doi.org/10.1007/s42835-024-01971-z","RegionNum":4,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
引用次数: 0

Abstract

Existing artificial neural network-based methodologies for salient object detection in RGB-depth (RGB-D) images typically require significant memory and computation time. In this paper, we propose ReBiT-Net, an novel and resource-efficient network designed to addresses this issue. ReBiT-Net utilizes a mobile network for feature extraction and incorporates depth map quality to regulate the fusion of multi-modal features, resulting in top-to-bottom refinement of salient objects using salient information. Empirical evaluations conducted on five benchmark datasets demonstrate the superior performance of our model in terms of accuracy (achieving 334 frames per second for an input size of 320 \(\times\) 320) and model parameters (merely 5.1 MB). Moreover, we introduce ReBiT-Net*, a simplified variant of ReBiT-Net, which entails reduced model parameters (4.2 MB) and enhanced processing speed (793 frames per second for a 256 \(\times\) 256 input size). These improvements are achieved through reduced memory requirements and computational demands via the adoption of a smaller input image size.

Abstract Image

查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
ReBiT-Net:用于 RGB-D 突出物体检测的资源节约型双向传输网络
现有的基于人工神经网络的 RGB 深度(RGB-D)图像中突出物体检测方法通常需要大量内存和计算时间。在本文中,我们提出了 ReBiT-Net,这是一种新颖且节省资源的网络,旨在解决这一问题。ReBiT-Net 利用移动网络进行特征提取,并结合深度图质量来调节多模态特征的融合,从而利用突出信息对突出对象进行从上到下的细化。在五个基准数据集上进行的实证评估表明,我们的模型在准确性(输入大小为 320 \(\times\) 320 时达到每秒 334 帧)和模型参数(仅为 5.1 MB)方面表现出色。此外,我们还引入了 ReBiT-Net* 这个 ReBiT-Net 的简化变体,它减少了模型参数(4.2 MB),提高了处理速度(256 (次) 256 输入大小每秒可处理 793 帧)。这些改进是通过采用较小的输入图像尺寸来减少内存需求和计算需求实现的。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 去求助
来源期刊
Journal of Electrical Engineering & Technology
Journal of Electrical Engineering & Technology ENGINEERING, ELECTRICAL & ELECTRONIC-
CiteScore
4.00
自引率
15.80%
发文量
321
审稿时长
3.8 months
期刊介绍: ournal of Electrical Engineering and Technology (JEET), which is the official publication of the Korean Institute of Electrical Engineers (KIEE) being published bimonthly, released the first issue in March 2006.The journal is open to submission from scholars and experts in the wide areas of electrical engineering technologies. The scope of the journal includes all issues in the field of Electrical Engineering and Technology. Included are techniques for electrical power engineering, electrical machinery and energy conversion systems, electrophysics and applications, information and controls.
期刊最新文献
Parameter Solution of Fractional Order PID Controller for Home Ventilator Based on Genetic-Ant Colony Algorithm Fault Detection of Flexible DC Grid Based on Empirical Wavelet Transform and WOA-CNN Aggregation and Bidding Strategy of Virtual Power Plant Power Management of Hybrid System Using Coronavirus Herd Immunity Optimizer Algorithm A Review on Power System Security Issues in the High Renewable Energy Penetration Environment
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1