資源描述:
《Exemplars for Object Detection.pdf》由會員上傳分享,免費在線閱讀,更多相關(guān)內(nèi)容在學(xué)術(shù)論文-天天文庫。
1、ExemplarsforObjectDetectionNoahSnavelyCS7670:September5,2011Announcements?Officehours:Thursdays1pm–2:30pm?CoursescheduleisnowonlineObjectdetection:wherearewe?Credit:Flickruserneilalderney123?Incredibleprogressinthelasttenyears?Betterfeatures,bettermode
2、ls,betterlearningmethods,betterdatasets?CombinationofscienceandhacksThe800-lbGorillaofVisionContests?PASCALVOCChallenge?20categories?Annualclassification,detection,segmentation,…challengesObjectdetectionperformance(2010)Objectdetectionperformance(2010)
3、The2011serveropenedforsubmissionstoday!Machinelearningforobjectdetection?Whatfeaturesdoweuse?–intensity,color,gradientinformation,…?Whichmachinelearningmethods?–generativevs.discriminative–k-nearestneighbors,boosting,SVMs,…?Whathacksdoweneedtogetthings
4、working?HistogramofOrientedGradients(HoG)HoGify10x10cells20x20cells[DalalandTriggs,CVPR2005]HistogramofOrientedGradients(HoG)HistogramofOrientedGradients(HoG)?LikeSIFT(ScaleInvariantFeatureTransform),but…–Sampledonadense,regulargrid–Gradientsarecontras
5、tnormalizedinoverlappingblocksHoGify10x10cells[DalalandTriggs,CVPR2005]20x20cellsHistogramofOrientedGradients(HoG)?Firstusedforapplicationofpersondetection[DalalandTriggs,CVPR2005]?CitedsinceinthousandsofcomputervisionpapersLinearclassifiers?Findlinear
6、functiontoseparatepositiveandnegativeexamplesxpositive:x?w?b?0iixnegative:x?w?b?0iiWhichlineisbest?[slidecredit:KristinGrauman]SupportVectorMachines(SVMs)?Discriminativeclassifierbasedonoptimalseparatingline(for2Dcase)?Maximizethemarginbetweenthepositi
7、veandnegativetrainingexamples[slidecredit:KristinGrauman]Supportvectormachines?Wantlinethatmaximizesthemargin.xpositive(y?1):x?w?b?1iiixnegative(y??1):x?w?b??1iiiForsupport,vectors,x?w?b??1iSupportvectorsMarginC.Burges,ATutorialonSupportVectorMachinesf
8、orPatternRecognition,DataMiningandKnowledgeDiscovery,1998[slidecredit:KristinGrauman]Persondetection,ca.20051.Representeachexamplewithasingle,fixedHoGtemplate2.Learnasingle[linear]SVMasadetectorCodeavailable:http://pascal.inrialpes.fr/s