Please use this identifier to cite or link to this item: http://repositorio.unicamp.br/jspui/handle/REPOSIP/327849
Full metadata record
DC FieldValueLanguage
dc.contributor.CRUESPUNIVERSIDADE DE ESTADUAL DE CAMPINASpt_BR
dc.identifier.isbn978-1-4673-9961-6pt
dc.typeCongressopt_BR
dc.titleDeep Neural Networks Underen
dc.contributor.authorCarvalhopt_BR
dc.contributor.authorMicael; Cordpt_BR
dc.contributor.authorMatthieu; Avilapt_BR
dc.contributor.authorSandra; Thomept_BR
dc.contributor.authorNicolas; Vallept_BR
dc.contributor.authorEduardopt_BR
unicamp.authorValle, Eduardo] Univ Estadual Campinas, UNICAMP, FEEC, DCA,RECOD Lab, Campinas, SP, Brazilpt_BR
unicamp.author.external[Carvalho, Micaelpt_BR
unicamp.author.externalCord, Matthieupt_BR
unicamp.author.externalThome, Nicolas] UPMC Univ Paris 06, Sorbonne Univ, CARS, LIP6 UMR 7606, 4 Pl Jussieu, F-75005 Paris, Francept_BR
unicamp.author.external[Carvalho, Micaelpt_BR
unicamp.author.externalAvila, Sandrapt_BR
dc.subjectFeature Robustnessen
dc.subjectDeep Teamingen
dc.subjectTransfer Learningen
dc.subjectImage Classificationen
dc.subjectFeature Compressionen
dc.description.abstractIn recent years, deep architectures have been used for transfer learning with state-of-the-art performance in many data sets. The properties of their features remain, however, largely unstudied under the transfer perspective. In this work, we present-an extensive analysis of the resiliency of feature vectors extracted front deep models, with special focus on the trade-off between performance and compression rate. By introducing perturbations to image descriptions extracted from a deep convolutional neural network, we change their precision and number of dimensions, measuring how it affects the final score. We show that deep features are more robust to these disturbances when compared to classical approaches, achieving a compression rate of 98.4%, while losing only 0.88% of their original score for Pascal VOC 2007.en
dc.relation.ispartof2016 IEEE International Conference on Image Processing (ICIP)pt_BR
dc.publisherIEEEpt_BR
dc.publisherNew Yorkpt_BR
dc.date.issued2016pt_BR
dc.identifier.citation2016 Ieee International Conference On Image Processing (icip). Ieee, p. 4443 - 4447, 2016.pt_BR
dc.language.isoEnglishpt_BR
dc.description.firstpage4443pt_BR
dc.description.lastpage4447pt_BR
dc.rightsfechadopt_BR
dc.sourceWOSpt_BR
dc.identifier.issn1522-4880pt_BR
dc.identifier.wosidWOS:000390782004083pt_BR
dc.identifier.urlhttp://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7533200pt_BR
dc.date.available2017-11-13T13:22:15Z-
dc.date.accessioned2017-11-13T13:22:15Z-
dc.description.provenanceMade available in DSpace on 2017-11-13T13:22:15Z (GMT). No. of bitstreams: 1 000390782004083.pdf: 1062187 bytes, checksum: cc4e4b5b6b3ccdb74612ace6eff40a61 (MD5) Previous issue date: 2016en
dc.identifier.urihttp://repositorio.unicamp.br/jspui/handle/REPOSIP/327849-
dc.description.conferencenome23rd IEEE International Conference on Image Processing (ICIP)pt_BR
dc.description.conferencedateSep 25-28, 2016pt_BR
dc.description.conferencelocationPhoenix, AZpt_BR
Appears in Collections:Unicamp - Artigos e Outros Documentos

Files in This Item:
File SizeFormat 
000390782004083.pdf1.04 MBAdobe PDFView/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.