OPTIMOL: automatic Object Picture collecTion via Incremental MOdel Learninga chicken and egg problem…Slide 3FrameworkSlide 5Slide 6Nonparametric topic model -Hierarchical Dirichlet Process (HDP)Slide 8ClassificationAnnotationPitfall #1: model driftSlide 12The “cache set”Slide 14ResultSlide 16Slide 17Slide 18OPTIMOL: automatic Object Picture collecTion via Incremental MOdel Learninga chicken and egg problem……among users, researchers, and dataImagesImagese.g. Caltech101,LabelMe, LHIFrameworkCategory model ClassificationDatasetKeyword: accordionLi, Wang & Fei-Fei, CVPR 2007FrameworkCategory model ClassificationDatasetKeyword: accordionLi, Wang & Fei-Fei, CVPR 2007…Compute SIFT descriptor [Lowe’99]Image representationImage representationKadir&Brady interest point detectorCodewords representationNonparametric topic model-Hierarchical Dirichlet Process (HDP)MNEach patchEach imageTeh, et al. 2004; Sudderth et al. CVPR 2006; Wang, Zhang & Fei-Fei, CVPR 2006Nonparametric topic model-Hierarchical Dirichlet Process (HDP)MNTeh, et al. 2004; Sudderth et al. CVPR 2006; Wang, Zhang & Fei-Fei, CVPR 2006Li, Wang & Fei-Fei, CVPR 2007ClassificationLikelihood ratio for decision: Category likelihood for I:Li, Wang & Fei-Fei, CVPR 2007AnnotationPitfall #1: model driftObject Model…Object Model…Li, Wang & Fei-Fei, CVPR 2007Object ModelGood ImagesBad Images…Pitfall #2: model diversityLi, Wang & Fei-Fei, CVPR 2007The “cache set”Li, Wang & Fei-Fei, CVPR 2007CategoryModelclassificationEnlarged datasetCacheRaw image datasetIncrementallearningResultLi, Wang & Fei-Fei, CVPR 2007Li, Wang & Fei-Fei, CVPR 2007Li, Wang & Fei-Fei, CVPR 2007OPTIMOL also learns good modelsTeam OPTIMOL (UIUC-Princeton): 11stst Place Place in the Software
View Full Document
Unlocking...