通过轻量级多流交叉和知识蒸馏实现高效 RGB-D 室内场景解析解决方案

IF 8.7 1区 工程技术 Q1 ENGINEERING, ELECTRICAL & ELECTRONIC IEEE Journal of Selected Topics in Signal Processing Pub Date : 2024-03-13 DOI:10.1109/JSTSP.2024.3400030
Wujie Zhou;Yuming Zhang;Weiqing Yan;Lv Ye
{"title":"通过轻量级多流交叉和知识蒸馏实现高效 RGB-D 室内场景解析解决方案","authors":"Wujie Zhou;Yuming Zhang;Weiqing Yan;Lv Ye","doi":"10.1109/JSTSP.2024.3400030","DOIUrl":null,"url":null,"abstract":"The rapid progression of convolutional neural networks (CNNs) has significantly improved indoor scene parsing, transforming the fields of robotics, autonomous navigation, augmented reality, and surveillance. Currently, societal demand is propelling these technologies toward integration into mobile smart device applications. However, the processing capabilities of mobile devices cannot support the comprehensive system requirements of CNNs, which poses a challenge for several deep-learning applications. One promising solution to this predicament is the deployment of lightweight student networks. These streamlined networks learn from their robust, cloud-based counterparts—that is, teacher networks—through knowledge distillation (KD). This facilitates a reduction in parameter count and optimizes student classification. Furthermore, a lightweight multiflow intersection network (LMINet) is proposed and developed for red–green–blue–depth (RGB-D) indoor scene parsing. The proposed method relies on dual-frequency KD (FKD) and compression KD (CKD) methods. A multiflow intersection module is introduced to efficiently integrate feature information from disparate layers. To maximize the performance of lightweight LMINet student (LMINet-S) networks, the FKD module employs a discrete cosine transform to capture feature information from different frequencies, whereas the CKD module compresses the features of diverse layers and distills their corresponding dimensions. Experiments using the NYUDv2 and SUN-RGBD datasets demonstrate that our LMINet teacher (LMINet-T) model, LMINet-S (without KD), and LMINet-S* (LMINet-S with KD) outperform state-of-the-art scene-parsing tools without increasing the parameter count (26.2M). Consequently, the technology is now closer to integration into mobile devices.","PeriodicalId":13038,"journal":{"name":"IEEE Journal of Selected Topics in Signal Processing","volume":null,"pages":null},"PeriodicalIF":8.7000,"publicationDate":"2024-03-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"An Efficient RGB-D Indoor Scene-Parsing Solution via Lightweight Multiflow Intersection and Knowledge Distillation\",\"authors\":\"Wujie Zhou;Yuming Zhang;Weiqing Yan;Lv Ye\",\"doi\":\"10.1109/JSTSP.2024.3400030\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"The rapid progression of convolutional neural networks (CNNs) has significantly improved indoor scene parsing, transforming the fields of robotics, autonomous navigation, augmented reality, and surveillance. Currently, societal demand is propelling these technologies toward integration into mobile smart device applications. However, the processing capabilities of mobile devices cannot support the comprehensive system requirements of CNNs, which poses a challenge for several deep-learning applications. One promising solution to this predicament is the deployment of lightweight student networks. These streamlined networks learn from their robust, cloud-based counterparts—that is, teacher networks—through knowledge distillation (KD). This facilitates a reduction in parameter count and optimizes student classification. Furthermore, a lightweight multiflow intersection network (LMINet) is proposed and developed for red–green–blue–depth (RGB-D) indoor scene parsing. The proposed method relies on dual-frequency KD (FKD) and compression KD (CKD) methods. A multiflow intersection module is introduced to efficiently integrate feature information from disparate layers. To maximize the performance of lightweight LMINet student (LMINet-S) networks, the FKD module employs a discrete cosine transform to capture feature information from different frequencies, whereas the CKD module compresses the features of diverse layers and distills their corresponding dimensions. Experiments using the NYUDv2 and SUN-RGBD datasets demonstrate that our LMINet teacher (LMINet-T) model, LMINet-S (without KD), and LMINet-S* (LMINet-S with KD) outperform state-of-the-art scene-parsing tools without increasing the parameter count (26.2M). Consequently, the technology is now closer to integration into mobile devices.\",\"PeriodicalId\":13038,\"journal\":{\"name\":\"IEEE Journal of Selected Topics in Signal Processing\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":8.7000,\"publicationDate\":\"2024-03-13\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"IEEE Journal of Selected Topics in Signal Processing\",\"FirstCategoryId\":\"5\",\"ListUrlMain\":\"https://ieeexplore.ieee.org/document/10529499/\",\"RegionNum\":1,\"RegionCategory\":\"工程技术\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"ENGINEERING, ELECTRICAL & ELECTRONIC\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"IEEE Journal of Selected Topics in Signal Processing","FirstCategoryId":"5","ListUrlMain":"https://ieeexplore.ieee.org/document/10529499/","RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"ENGINEERING, ELECTRICAL & ELECTRONIC","Score":null,"Total":0}
引用次数: 0

摘要

卷积神经网络(CNN)的飞速发展极大地改进了室内场景解析,改变了机器人、自主导航、增强现实和监控领域。目前,社会需求正推动这些技术集成到移动智能设备应用中。然而,移动设备的处理能力无法支持 CNN 的全面系统要求,这给一些深度学习应用带来了挑战。解决这一困境的一个可行方案是部署轻量级学生网络。这些精简的网络通过知识提炼(KD)从其基于云的强大同类网络(即教师网络)中学习。这有助于减少参数数量,优化学生分类。此外,针对红-绿-蓝-深(RGB-D)室内场景解析,提出并开发了一种轻量级多流交叉网络(LMINet)。所提出的方法依赖于双频 KD(FKD)和压缩 KD(CKD)方法。该方法引入了多流交叉模块,以有效整合来自不同层的特征信息。为了最大限度地提高轻量级 LMINet 学生(LMINet-S)网络的性能,FKD 模块采用离散余弦变换来捕捉不同频率的特征信息,而 CKD 模块则压缩不同层的特征并提炼其相应维度。使用 NYUDv2 和 SUN-RGBD 数据集进行的实验表明,我们的 LMINet 教师(LMINet-T)模型、LMINet-S(不含 KD)和 LMINet-S*(含 KD 的 LMINet-S)在不增加参数数(26.2M)的情况下,性能优于最先进的场景解析工具。因此,该技术现在更接近于集成到移动设备中。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
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
An Efficient RGB-D Indoor Scene-Parsing Solution via Lightweight Multiflow Intersection and Knowledge Distillation
The rapid progression of convolutional neural networks (CNNs) has significantly improved indoor scene parsing, transforming the fields of robotics, autonomous navigation, augmented reality, and surveillance. Currently, societal demand is propelling these technologies toward integration into mobile smart device applications. However, the processing capabilities of mobile devices cannot support the comprehensive system requirements of CNNs, which poses a challenge for several deep-learning applications. One promising solution to this predicament is the deployment of lightweight student networks. These streamlined networks learn from their robust, cloud-based counterparts—that is, teacher networks—through knowledge distillation (KD). This facilitates a reduction in parameter count and optimizes student classification. Furthermore, a lightweight multiflow intersection network (LMINet) is proposed and developed for red–green–blue–depth (RGB-D) indoor scene parsing. The proposed method relies on dual-frequency KD (FKD) and compression KD (CKD) methods. A multiflow intersection module is introduced to efficiently integrate feature information from disparate layers. To maximize the performance of lightweight LMINet student (LMINet-S) networks, the FKD module employs a discrete cosine transform to capture feature information from different frequencies, whereas the CKD module compresses the features of diverse layers and distills their corresponding dimensions. Experiments using the NYUDv2 and SUN-RGBD datasets demonstrate that our LMINet teacher (LMINet-T) model, LMINet-S (without KD), and LMINet-S* (LMINet-S with KD) outperform state-of-the-art scene-parsing tools without increasing the parameter count (26.2M). Consequently, the technology is now closer to integration into mobile devices.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
IEEE Journal of Selected Topics in Signal Processing
IEEE Journal of Selected Topics in Signal Processing 工程技术-工程:电子与电气
CiteScore
19.00
自引率
1.30%
发文量
135
审稿时长
3 months
期刊介绍: The IEEE Journal of Selected Topics in Signal Processing (JSTSP) focuses on the Field of Interest of the IEEE Signal Processing Society, which encompasses the theory and application of various signal processing techniques. These techniques include filtering, coding, transmitting, estimating, detecting, analyzing, recognizing, synthesizing, recording, and reproducing signals using digital or analog devices. The term "signal" covers a wide range of data types, including audio, video, speech, image, communication, geophysical, sonar, radar, medical, musical, and others. The journal format allows for in-depth exploration of signal processing topics, enabling the Society to cover both established and emerging areas. This includes interdisciplinary fields such as biomedical engineering and language processing, as well as areas not traditionally associated with engineering.
期刊最新文献
Front Cover Table of Contents IEEE Signal Processing Society Information Introduction to the Special Issue Near-Field Signal Processing: Algorithms, Implementations and Applications IEEE Signal Processing Society Information
×
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