Compressing Deep CNNs Using Basis Representation and Spectral Fine-Tuning
We propose an efficient and straightforward method for compressing deep convolutional neural networks (CNNs) that uses basis filters to represent the convolutional layers, and optimizes the performance of the compressed network directly in the basis space. Specifically, any spatial convolution layer...
Saved in:
Published in | 2021 IEEE International Conference on Image Processing (ICIP) pp. 3537 - 3541 |
---|---|
Main Authors | , , |
Format | Conference Proceeding |
Language | English |
Published |
IEEE
19.09.2021
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Abstract | We propose an efficient and straightforward method for compressing deep convolutional neural networks (CNNs) that uses basis filters to represent the convolutional layers, and optimizes the performance of the compressed network directly in the basis space. Specifically, any spatial convolution layer of the CNN can be replaced by two successive convolution layers: the first is a set of three-dimensional orthonormal basis filters, followed by a layer of one-dimensional filters that represents the original spatial filters in the basis space. We jointly fine-tune both the basis and the filter representation to directly mitigate any performance loss due to the truncation. Generality of the proposed approach is demonstrated by applying it to several well known deep CNN architectures and data sets for image classification and object detection. We also present the execution time and power usage at different compression levels on the Xavier Jetson AGX processor. |
---|---|
AbstractList | We propose an efficient and straightforward method for compressing deep convolutional neural networks (CNNs) that uses basis filters to represent the convolutional layers, and optimizes the performance of the compressed network directly in the basis space. Specifically, any spatial convolution layer of the CNN can be replaced by two successive convolution layers: the first is a set of three-dimensional orthonormal basis filters, followed by a layer of one-dimensional filters that represents the original spatial filters in the basis space. We jointly fine-tune both the basis and the filter representation to directly mitigate any performance loss due to the truncation. Generality of the proposed approach is demonstrated by applying it to several well known deep CNN architectures and data sets for image classification and object detection. We also present the execution time and power usage at different compression levels on the Xavier Jetson AGX processor. |
Author | Mahalanobis, Abhijit Khan, Fahad Ahmad Tayyab, Muhammad |
Author_xml | – sequence: 1 givenname: Muhammad surname: Tayyab fullname: Tayyab, Muhammad organization: University of Central Florida,Center for Research in Computer Vision,Department of Computer Science,USA – sequence: 2 givenname: Fahad Ahmad surname: Khan fullname: Khan, Fahad Ahmad organization: University of Central Florida,Center for Research in Computer Vision,Department of Computer Science,USA – sequence: 3 givenname: Abhijit surname: Mahalanobis fullname: Mahalanobis, Abhijit organization: University of Central Florida,Center for Research in Computer Vision,Department of Computer Science,USA |
BookMark | eNotz91KAzEQBeAoCtrqEwiSF9g6M_lxc6mr1YVSRdvrkt1MJdKmy2a98O2t2qvDgY8DZyRO0i6xENcIE0RwN3VVv2pyVE4ICCfOgEUqj8QIrTVaIxo8FuekSixKo92ZGOX8CUCACs9FXe22Xc85x_QhH5g7Wc3nWS7_-r3PMcs3_gWcBj_EXZI-BfnecTv0fiOnMXGx-Ep7fSFO136T-fKQY7GcPi6q52L28lRXd7MiEqihwDY4cIQajCbVBgpswDUe3K3VSKFtbAt70ZBTtgnsXevsGoNuSBljSjUWV_-7kZlXXR-3vv9eHW6rH6-RTaY |
ContentType | Conference Proceeding |
DBID | 6IE 6IH CBEJK RIE RIO |
DOI | 10.1109/ICIP42928.2021.9506128 |
DatabaseName | IEEE Electronic Library (IEL) Conference Proceedings IEEE Proceedings Order Plan (POP) 1998-present by volume IEEE Xplore All Conference Proceedings IEEE Xplore IEEE Proceedings Order Plans (POP) 1998-present |
DatabaseTitleList | |
Database_xml | – sequence: 1 dbid: RIE name: IEEE/IET Electronic Library (IEL) url: https://proxy.k.utb.cz/login?url=https://ieeexplore.ieee.org/ sourceTypes: Publisher |
DeliveryMethod | fulltext_linktorsrc |
Discipline | Applied Sciences |
EISBN | 1665441151 9781665441155 |
EISSN | 2381-8549 |
EndPage | 3541 |
ExternalDocumentID | 9506128 |
Genre | orig-research |
GroupedDBID | 6IE 6IF 6IH 6IK 6IL 6IM 6IN AAJGR ADZIZ ALMA_UNASSIGNED_HOLDINGS BEFXN BFFAM BGNUA BKEBE BPEOZ CBEJK CHZPO IEGSK IJVOP IPLJI JC5 M43 OCL RIE RIL RIO RNS |
ID | FETCH-LOGICAL-i203t-1cd90921405423cd2de509ba0976412dcb6c0909b2936bdea9c96f1d4b2355583 |
IEDL.DBID | RIE |
IngestDate | Wed Jun 26 19:30:59 EDT 2024 |
IsPeerReviewed | false |
IsScholarly | true |
Language | English |
LinkModel | DirectLink |
MergedId | FETCHMERGED-LOGICAL-i203t-1cd90921405423cd2de509ba0976412dcb6c0909b2936bdea9c96f1d4b2355583 |
PageCount | 5 |
ParticipantIDs | ieee_primary_9506128 |
PublicationCentury | 2000 |
PublicationDate | 2021-Sept.-19 |
PublicationDateYYYYMMDD | 2021-09-19 |
PublicationDate_xml | – month: 09 year: 2021 text: 2021-Sept.-19 day: 19 |
PublicationDecade | 2020 |
PublicationTitle | 2021 IEEE International Conference on Image Processing (ICIP) |
PublicationTitleAbbrev | ICIP |
PublicationYear | 2021 |
Publisher | IEEE |
Publisher_xml | – name: IEEE |
SSID | ssj0020131 |
Score | 2.212173 |
Snippet | We propose an efficient and straightforward method for compressing deep convolutional neural networks (CNNs) that uses basis filters to represent the... |
SourceID | ieee |
SourceType | Publisher |
StartPage | 3537 |
SubjectTerms | Basis representation Conferences Convolution Convolutional neural networks Image classification Image coding network compression Object detection orthogonal filters Spatial filters |
Title | Compressing Deep CNNs Using Basis Representation and Spectral Fine-Tuning |
URI | https://ieeexplore.ieee.org/document/9506128 |
hasFullText | 1 |
inHoldings | 1 |
isFullTextHit | |
isPrint | |
link | http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwjV3PT8IwFH4BTp5Qwfg7PXi0Y-u2br2KEjCBEAMJN9KuD0M0g7Dt4l9vWyZG48HbsqxZ05fm-177fe8B3HGNOhJJSIWUnEbIImpAOKQMpURDH3gqrFF4POHDefS8iBcNuD94YRDRic_Qs4_uLl9vssoelfVEbAE5bUIz9dneq3VIrmzdmNoBHPiiN-qPprYTk1VvscCrR_5ooeIQZNCG8de_98KRN68qlZd9_CrL-N_JHUP326tHpgcUOoEG5qfQrsklqbdu0YGR3flO9Jq_kkfELelPJgVxkgHyIIt1QV6cKrY2I-VE5ma8NWLu5DsZGDZKZ5U9RenCfPA06w9p3UeBrpkfljTItPAFM6lUbMhTpplGQxOU9A0ViQKmM8Uz33yhTJC40ihFJvgq0JFioS0HFp5BK9_keA5EMp5wqU3itsJISRQrLlORxLE2TEKH6gI6dmWW232pjGW9KJd_v76CIxsdK78IxDW0yl2FNwbjS3XrgvsJDNGluQ |
link.rule.ids | 310,311,786,790,795,796,802,23958,23959,25170,27958,55109 |
linkProvider | IEEE |
linkToHtml | http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwjV3PT8IwFH5BPOgJFYy_7cGjnVu3FXoVJUxhIQYSbqRdH4ZoBmHbxb_edkyMxoO3ZWuzpS_N973u-94DuOEadSDaPhVSchogC6gBYZ8ylBINfeAdYY3Cw5j3J8HTNJzW4HbrhUHEUnyGjr0s_-XrZVLYo7I7EVpA7uzArsF5V2zcWtv0ylaOqTzA5uFd1I1GtheT1W8xz6nm_miiUmJIrwHDr7dvpCNvTpErJ_n4VZjxv593AK1vtx4ZbXHoEGqYHkGjopek2rxZEyK790vZa_pKHhBXpBvHGSlFA-ReZouMvJS62MqOlBKZmvnWirmW76Rn-CgdF_YcpQWT3uO426dVJwW6YK6fUy_RwhXMJFOhoU-JZhoNUVDSNWQk8JhOFE9cM0KZMHGlUYpE8LmnA8V8WxDMP4Z6ukzxBIhkvM2lNqnbHAMlUcy57Ih2GGrDJbSvTqFpV2a22hTLmFWLcvb37WvY64-Hg9kgip_PYd9GyooxPHEB9Xxd4KVB_FxdlYH-BPzaqQ8 |
openUrl | ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=proceeding&rft.title=2021+IEEE+International+Conference+on+Image+Processing+%28ICIP%29&rft.atitle=Compressing+Deep+CNNs+Using+Basis+Representation+and+Spectral+Fine-Tuning&rft.au=Tayyab%2C+Muhammad&rft.au=Khan%2C+Fahad+Ahmad&rft.au=Mahalanobis%2C+Abhijit&rft.date=2021-09-19&rft.pub=IEEE&rft.eissn=2381-8549&rft.spage=3537&rft.epage=3541&rft_id=info:doi/10.1109%2FICIP42928.2021.9506128&rft.externalDocID=9506128 |