Cet ouvrage et des milliers d'autres font partie de la bibliothèque YouScribe
Obtenez un accès à la bibliothèque pour les lire en ligne
En savoir plus

Partagez cette publication

DatasetIssuesinObjectRecognition1J.Ponce1,2,T.L.Berg3,M.Everingham4,D.A.Forsyth1,M.Hebert5,S.Lazebnik1,M.Marszalek6,C.Schmid6,B.C.Russell7,A.Torralba7,C.K.I.Williams8,J.Zhang6,andA.Zisserman4UniversityofIllinoisatUrbana-Champaign,USA2EcoleNormaleSup´erieure,Paris,France3UniversityofCaliforniaatBerkeley,USA4OxfordUniversity,UKCarnegieMellonUniversity,Pittsburgh,USAINRIARhoˆne-Alpes,Grenoble,France7MIT,Cambridge,USAUniversityofEdinburgh,Edinburgh,UK586Abstract.Appropriatedatasetsarerequiredatallstagesofobjectrecognitionresearch,includinglearningvisualmodelsofobjectandscenecategories,detectingandlocalizinginstancesofthesemodelsinim-ages,andevaluatingtheperformanceofrecognitionalgorithms.Currentdatasetsarelackinginseveralrespects,andthispaperdiscussessomeofthelessonslearnedfromexistingefforts,aswellasinnovativewaystoobtainverylargeanddiverseannotateddatasets.Italsosuggestsafewcriteriaforgatheringfuturedatasets.1IntroductionImagedatabasesareanessentialelementofobjectrecognitionresearch.Theyarerequiredforlearningvisualobjectmodelsandfortestingtheperformanceofclassification,detection,andlocalizationalgorithms.Infact,publiclyavailableimagecollectionssuchasUIUC[1],Caltech4[10],andCaltech101[9]haveplayedakeyroleintherecentresurgenceofcategory-levelrecognitionresearch,drivingthefieldbyprovidingacommongroundforalgorithmdevelopmentandevaluation.Currentdatasets,however,offerasomewhatlimitedrangeofimagevariability:Althoughtheappearance(andtosomeextent,theshape)ofobjectsdoesindeedvarywithineachclass(e.g.,amongtheairplanes,cars,faces,andmotorbikesofCaltech4),theviewpointsandorientationsofdifferentinstancesineachcategorytendtobesimilar(e.g.,sideviewsofcarstakenbyahorizontalcamerainUIUC);theirsizesandimagepositionsarenormalized(e.g.,theobjectsofinteresttakeupmostoftheimageandareapproximatelycenteredinCaltech101);thereisonlyoneinstanceofanobjectperimage;finally,thereislittleornoocclusionandbackgroundclutter.ThisisillustratedbyFigures1and3fortheCaltech101database,butremainstrueofmostdatasetsavailabletoday.Theproblemswithsuchrestrictionsaretwofold:(i)somealgorithmsmayexploitthem(forexamplenear-globaldescriptorswithnoscaleorrotationin-variancemayperformwellonsuchimages),yetwillfailwhentherestrictions
2Fi.g.1SampleimagesrfmoehtCaltech101dataset,]9[courtesyfo-ieFieFL.i
Un pour Un
Permettre à tous d'accéder à la lecture
Pour chaque accès à la bibliothèque, YouScribe donne un accès à une personne dans le besoin