Combined Object Categorization and Segmentation with an Implicit Shape Model B. Leibe, A. Leonardis, and B. Schiele. Addressed ProblemsAddressed ProblemsAddressed ProblemsProposed solutionRec./Seg. procedureConstruction of the codebookConstruction of prob. distributionProbabilistic framework(1/2)Probabilistic framework(2/2)Experimental results: carsExperimental results: carsExperimental results: cowsObject Segmentation (1/3)Object Segmentation (2/3)Segmentation results (3/3)Hypothesis verification (1/2)Hypothesis verification (2/2)Experimental results: carsComparisonOpen QuestionsReferencesCombined Object Categorization and Segmentation with an Implicit Shape ModelB. Leibe, A. Leonardis, and B. Schiele. Marco [email protected]/CNS 150Advanced Topics in Vision.Visual RecognitionAddressed ProblemsCategory detection and localization (cars)Addressed Problems“Soft” segmentationAddressed ProblemsMultiple-Object Scene Analysis Recognition of Articulated Object (???)Proposed solution• Learning an– codebook of local patches– Non parametric prob. distribution of a patch position (w.r.t. the center)Similar to the solution of S. Agarwal and D. Roth 2002Rec./Seg. procedureConstruction of the codebookConstruction of prob. distributionl , l , .....21Non parametric density estimation: kernel implicitly definedSegmented training dataProbabilistic framework(1/2)Probabilistic framework(2/2)Mean Shift Algorithm for local maximaExperimental results: cars50x2 training examplesExperimental results: carsExperimental results: cowsObject Segmentation (1/3)• GivenObject Segmentation (2/3)Segmentation results (3/3)Hypothesis verification (1/2)MDL principle (Occam's razor)Hypothesis verification (2/2)Combinatorial complexity of space HExperimental results: carsComparisonUICI car databaseOpen Questions• Large training set for non-parametric est.• Segmented training set• Shape partially modeled• Not too articulated objects• Multiple classes in one image?• Star model?References• [1] B. Leibe, A. Leonardis, and B. Schiele. Combined Object Categorization and Segmentation with an Implicit Shape Model. in ECCV'04 Workshop on Statistical Learning in Computer Vision, Prague, May 2004.• [2] B. Leibe. Interleaved Object Categorization and Segmentation. Doctor of Technical Sciences Thesis. Diss. ETH no15752, Zurich 2004• [3] D. Comaniciu, and P. Meer. Mean Shift Analysis and applications. IEEE Int. Conf. Computer Vision (ICCV'99), Kerkyra, Greece, 1197-1203,
View Full Document