Venkatesh Babu Senior Member IEEE Ravi Kiran Sarvadevabhatla ravikiransslserciiscin and R Venkatesh Babu venkyserciiscin are with Video Analytics Lab Supercomputer Education and Research Centre Indian Institute of Science Bangalore 560 012 India brP ID: 84156
Download Pdf The PPT/PDF document "Understanding categoryepitomes and epito..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
2 Fig.1:Medianepitome-scores(y-axis)andcorrespondingmedian-absolute-deviationbarsfor50objectcategories(x-axis). Fig.2:Category-epitomesforarm. Fig.3:Category-epitomesforcarrot.Fig.4:Theobservationsinthemainpaperare,tosomeextent,temperedbythevaryingrangeoferrorbars(Figure1)inepitome-scoresforsomeofthecategories.Thismakessensesincethenotionofobjectcategoryismeanttoencompassthevarietyinattributesoftheobjectsuchassize,shapeandappearance.Thereadyconclusion,then,isthatsomeobjectcategoriesmayhavemultiplerepresentativecategory-epitomesratherthanasingle,uniquerepresentative.Forexample,thesketchesofthecategory(human)armarelikelydrawnwithafairlyconsistentappearancebyhumans.Thisconsistencyinuencesthesketchclassierandcancauseittoproducecategory-epitomeswhichexhibitfairlyminorvariations(seeFigure2),leadingtoacompactdistributionofepitome-scores.Incontrast,thevarietyinsketchesofacategorysuchascarrotisreectedinthecorrespondingcategory-epitomes(seeFigure3)andbyextension,inthelargedeviationoftheepitome-scores. 3 Fig.5:Effectofbudgetingepitome-scoreoncategoryrecognitionrateusingdatafrom10selectedcategories. Fig.6:Category-epitomesforboomerang. Fig.7:Category-epitomesforbridge. Fig.8:Category-epitomesforbicycle.Fig.9:Figure5demonstratestheeffectofbudgetingepitome-scoreoncategoryrecognitionrateusingdatafrom10selectedcategories.Valuesofvariouspossibleepitome-scorethresholdsareshownonx-axis.The%ofimagesineachcategorywhichexceedaparticularvalueofthethresholdareshownony-axis.The10categorynamesareshownadjacenttotheirrespectiveplots.Epitome-scorecanbeconsideredasaproxyforsemanticlevel-of-detail.Viewedinthislight,theplotfromFigure5suggeststhevaryingepitome-scorebudgetsacrosscategories.Forexample,categoriessuchasbridgeandbicycle(plotstowardsthelowerrightinFigure5)requireaconsiderablelevelofdetailbeforetheirepitomalavatarsarerevealed(Figures7,8).Ontheotherhand,categorieswithplotstowardstheupperleftcorner(e.g.boomerang)inFigure5haverelativelylessstringentdemandsonlevelofdetail(seeFigure6).