Multi-taskFaceAttributeRecognitionBasedonUnbalancedClasses

doi:10.3969/j.issn.1006-2475.2018.06.012

Abstract

Abstract: Therecognitionofattributesplaysanimportantroleinobjectrecognition,suchasfaceverification,activityrecognitioninvideo.Theimprovementofattributerecognitionwillleadtobetterresultfortheapplicationwhichusestheseattributes.Inrecentyears,therearesomeworksfocusingontheattributelearning.However,attributeshavebeenconsideredtobeindependentinmostworks.Inpractice,weknowthatisnotrightandmanyattributesarestronglyrelated,suchasnobeardandfemale,baldandhaircolor.Inaddition,mostworksneglecttheunbalanceofdifferenceclasssamples,suchasimagesofbaldisaverysmallpartofsamples.Basedonthetwoobservations,weproposeamulti-taskunbalancedfacialattributerecognitionframeworkusingmodifiedDensenet.Toacertaindegrees,theframeworkrelievestheunbalancedproblem.Theproposedmethodoutperformsothermethods,haslessparametersandrunsfaster.Wedemonstratetheeffectivenessofourmethodontwochallengingpubliclyavailabledatasets.

Key words: faceattribute, deeplearning, multi-tasklearning, multi-labellearning

CLC Number:

TP391

ZHANGWen. Multi-taskFaceAttributeRecognitionBasedonUnbalancedClasses[J]. Computer and Modernization, doi: 10.3969/j.issn.1006-2475.2018.06.012.

References ［22］

［1］	DuanK,ParikhD,CrandallD,etal.Discoveringlocalizedattributesforfine-trainedrecognition［C］//2012IEEEConferenceonComputerVisionandPatternRecognition.2012:3474-3481.
［2］	PhillipsPJ,ZhengJingjing,JiangZhuolin,etal.Submodularattributeselectionforactionrecognitioninvideo［C］//AdvancesinNeuralInformationProcessingSystems.2014:1341-1349.
［3］	KumarN,BergAC,BelhumeurPN,etal.Attributeandsimileclassifiersforfaceverification［C］//IEEEInternationalConferenceonComputerVision(ICCV).2009:365-372.〖JP〗
［4］	KumarN,BergAC,BelhumeurPN,etal.Describablevisualattributesforfaceverificationandimagesearch［J］.IEEETransactionsonPatternAnalysisandMachineIntelligence,2011,33(10):1962-1977.
［5］	WuZhong,KeQifa,SunJan,etal.Scalablefaceimageretrievalwithidentity-basedquantizationandmultireferencereranking［J］.IEEETransactionsonPatternAnalysisandMachineIntelligence,2011,33(10):1991-2001.
［6］	LiuZiwei,LuoPing,WangXiaogang,etal.Deeplearningfaceattributesinthewild［C］//2015IEEEInternationalConferenceonComputerVision.2015:3730-3738.
［7］	ZhangNing,PaluriM,RanzatoM,etal.Panda:Posealignednetworksfordeepattributemodeling［C］//2014IEEEInternationalConferenceonComputerVision.2014:1637-1644.
［8］	KrizhevskyA,SutskeverI,HintonGE.ImageNetclassificationwithdeepconvolutionalneuralnetworks［C］//AdvancesinNeuralInformationProcessingSystems.2012:1097-1105.
［9］	HuangGao,LiuZhuang,WeinbergerKQ.Denselyconnectedconvolutionalnetworks［C］//2017IEEEConferenceonComputerVisionandPatternRecognition.2017:2261-2269.
［10］	AbrarHA,WangGang,LuJiwen,etal.Multi-taskCNNmodelforattributeprediction［J］.IEEETransactionsonMultimedia,2015,17(11):1949-1977.
［11］	LeviG,HassnerT.Ageandgenderclassificationusingconvolutionalneuralnetworks［C］//2015IEEEConferenceonComputerVisionandPatternRecognitionWorkshops.2015:34-42.
［12］	〖JP2〗RenShaoqing,HeKaiming,GirshickRB.FasterR-CNN:〖JP〗Towardsreal-timeobjectdetectionwithregionproposalnetworks［J］.IEEETransactionsonPatternAnalysisandMachineIntelligence,2015,39(6):1137-1149.
［13］	YangShuo,LuoPing,LoyCC,etal.Fromfacialpartsresponsestofacedetection:Adeeplearningapproach［C］//2015IEEEInternationalConferenceonComputerVision.2015:3676-3684.
［14］	SunYi,WangXiaogang,TangXiaoou.Deeplearningfacerepresentationfrompredicting10,000classes［C］//2014IEEEConferenceonComputerVisionandPatternRecognition.2014:1891-1898.
［15］	〖JP2〗DevriesT,BiswaranjanK,TaylorGW.Multi-tasklearningoffaciallandmarksandexpression［C］//2014CanadianConferenceonComputerandRobotVision.2014:98-103.〖JP〗
［16］	ZhouQiang,WangGang,JiaKui,etal.Learningtosharelatenttasksforactionrecognition［C］//2013IEEEInternationalConferenceonComputerVision.2013:2264-2271.
［17］	HuangYan，WangYan,WangLiang.Unconstrainedmultimodalmulti-labellearning［J］.IEEETransactionsonMultimedia,2015,17(11):1923-1935.
［18］	WangYang,MoriG.Adiscriminativelatentmodelofobjectclassesandattributes［C］//The11thEuropeanConferenceonComputerVision.2010:155-168.
［19］	RuddEM,GntherM,BoultTE.MOON:Amixedobjectiveoptimizationnetworkfortherecognitionoffacialattributes［C］//The14thEuropeanConferenceonComputerVision.2016:19-35.
［20］	HandEM,ChellappaR.Attributesforimprovedattributes:Amulti-tasknetworkutilizingimplicitandexplicitrelationshipsforfacialattributeclassification［C］//Proceedingsofthe31stAAAIConferenceonArtificialIntelligence.2017:4068-4074.
［21］	HuangGB,RameshM,BergT,etal.LabeledFacesintheWild:ADatabaseforStudyingFaceRecognitioninUnconstrainedEnvironments［R］.UniversityofMassachusettsAmherst,2007.
［22］	ZhongYang,SullivanJ,LiHaibo.Faceattributepredictionusingoff-the-shelfCNNfeatures［C］//2016InternationalConferenceonBiometrics.2016:1-7.

[1]	HEJun-jie,XIAOKe,LIUChang,CHENSong-yan. TFT-LCDCircuitDefectsDetectionBasedonFasterR-CNN [J]. Computer and Modernization, 2018, 0(07): 33-.
[2]	LINXue-feng1,XIAYuan-yi2,GUOJin-long1,YUXiao-wen1. SensitiveFileDetectionMethodBasedonCNN [J]. Computer and Modernization, 2018, 0(07): 28-.