DOC PREVIEW
UIUC CS 543 - Object Category Recognition

This preview shows page 1-2-3-4-30-31-32-33-34-61-62-63-64 out of 64 pages.

Save
View full document
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
View full document
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience
Premium Document
Do you want full access? Go Premium and unlock all 64 pages.
Access to all documents
Download any document
Ad free experience

Unformatted text preview:

Object Category RecognitionToday’s class: categorizationImage CategorizationMany classifiers to choose fromNo Free Lunch TheoremClassifiers: Linear SVMSlide 7Slide 8Classifiers: Kernelized SVMUsing SVMsClassifiers: Decision TreesBoosted Decision TreesUsing Boosted Decision TreesK-nearest neighbor1-nearest neighbor3-nearest neighbor5-nearest neighborUsing K-NNWhat to remember about classifiersSome Machine Learning ReferencesMoving on…Object category detection in computer visionGeneral Process of Object RecognitionSpecifying an object modelSlide 27Slide 28Slide 29Slide 30Generating hypothesesSlide 32Slide 33Slide 34Slide 35Resolving detection scoresSlide 37More thoughts on categories…Object CategorizationSlide 40An example of categorical perceptionCategorical perception: “sharp” boundariesContinuous perception: graded responseWhy do we care about categories?Slide 45The perception of functionDirect perceptionSlide 48Limitations of Direct PerceptionSlide 50So categorize or not?How many categories?ManyHow many object categories are there?Slide 55Which level of categorization is the right one?How do you define a category?Prototype or Sum of Exemplars ?Slide 59Levels of CategorizationRosch’s Levels of CategorizationSlide 62Typicality effectsEntry-level categories (Jolicoeur, Gluck, Kosslyn 1984)We do not need to recognize the exact categoryNext classObject Category RecognitionComputer VisionCS 543 / ECE 549 University of IllinoisDerek Hoiem03/16/10(Plus leftover material from image categorization)Today’s class: categorization•More about classifiers•Overview of object category detection•More about categorization in generalImage CategorizationTraining LabelsTraining ImagesClassifier TrainingTrainingImage FeaturesImage FeaturesTestingTest ImageTrained ClassifierTrained ClassifierOutdoorPredictionMany classifiers to choose from•SVM•Neural networks•Naïve Bayes•Bayesian network•Logistic regression•Randomized Forests•Boosted Decision Trees•K-nearest neighbor•RBMs•Etc.No Free Lunch TheoremClassifiers: Linear SVMx xxxxxxxooooox2x1Classifiers: Linear SVMx xxxxxxxooooox2x1Classifiers: Linear SVMx xxxxxxxoooooox2x1Classifiers: Kernelized SVMxx xx oo oxxxxxoooxx2Using SVMs•Good general purpose classifier–Generalization depends on margin, so works well with many weak features–No feature selection–Usually requires some parameter tuning•Choosing kernel–Linear: fast training/testing – start here–RBF: related to neural networks, nearest neighbor–Chi-squared, histogram intersection: good for histograms (but slower, esp. chi-squared)–Can learn a kernel functionClassifiers: Decision Treesx xxxxxxxooooooox2x1Boosted Decision Trees …Gray?High inImage?Many LongLines?YesNoNoNoNoYes YesYesVery High Vanishing Point?High in Image?Smooth? Green?Blue?YesNoNoNoNoYes YesYesGround Vertical Sky[Collins et al. 2002]P(label | good segment, data)Using Boosted Decision Trees•Flexible: can deal with both continuous and categorical variables•How to control bias/variance trade-of–Size of trees–Number of trees•Boosting trees often works best with a small number of well-designed features•Boosting “stubs” can give a fast classifierK-nearest neighborx xxxxxxxooooooox2x1++1-nearest neighborx xxxxxxxooooooox2x1++3-nearest neighborx xxxxxxxooooooox2x1++5-nearest neighborx xxxxxxxooooooox2x1++Using K-NN•Simple, so another good one to try first•With infinite examples, 1-NN provably has error that is at most twice Bayes optimal errorWhat to remember about classifiers•No free lunch: machine learning algorithms are tools, each well-suited to some purposes but not others•Try simple classifiers first•Better to have smart features and simple classifiers than simple features and smart classifiers•Use increasingly powerful classifiers with more training data (bias-variance tradeof)Some Machine Learning References•General–Tom Mitchell, Machine Learning, McGraw Hill, 1997–Christopher Bishop, Neural Networks for Pattern Recognition, Oxford University Press, 1995•Adaboost–Friedman, Hastie, and Tibshirani, “Additive logistic regression: a statistical view of boosting”, Annals of Statistics, 2000 •SVMs–http://www.support-vector.net/icml-tutorial.pdfMoving on…Object CategoriesObject category detection in computer visionGoal: detect all pedestrians, cars, monkeys, etc in imageGeneral Process of Object RecognitionSpecify Object ModelGenerate HypothesesScore HypothesesResolutionSpecifying an object model1. Statistical Template in Bounding Box–Object is some (x,y,w,h) in image–Features defined wrt bounding box coordinatesImage Template VisualizationImages from FelzenszwalbSpecifying an object model2. Articulated parts model–Object is configuration of parts–Each part is detectableImages from FelzenszwalbSpecifying an object model3. Hybrid template/parts modelDetectionsTemplate VisualizationFelzenszwalb et al. 2008Specifying an object model4. 3D-ish model•Object is collection of 3D planar patches under affine transformationGeneral Process of Object RecognitionSpecify Object ModelGenerate HypothesesScore HypothesesResolutionGenerating hypotheses1. Sliding window–Test patch at each location and scaleGenerating hypotheses1. Sliding window–Test patch at each location and scaleGenerating hypotheses2. Voting from patches/keypointsInterest PointsMatched Codebook EntriesProbabilistic Voting3D Voting Space(continuous)xysISM model by Leibe et al.Generating hypotheses3. Region-based proposalGeneral Process of Object RecognitionSpecify Object ModelGenerate HypothesesScore HypothesesResolutionMany types of classifiers, featuresResolving detection scores1. Non-max suppressionScore = 0.1Score = 0.8 Score = 0.8Resolving detection scores2. Context/reasoningmetersmetersMore thoughts on categories…Object CategorizationObject Categorization?An example of categorical perceptionContinuous perception: graded response50 100 150 200 25050100150200250Many perceptual phenomena are a mixture of the two: categorical at an everyday level of magnification, but continuous at a more microscopic level.50 100 150 200 25050100150200250Categorical perception: “sharp” boundariesSlide Credit: TorralbaCategorical perception: “sharp” boundarieshappinessfear•Identification Task% identifcationAngerFearHappiness50 100 150 200 25050100150200250Red vs. yellow vs. green vs. blueSlide Credit: TorralbaContinuous perception: graded response20-2425-2930-34 35-39


View Full Document
Download Object Category Recognition
Our administrator received your request to download this document. We will send you the file to your email shortly.
Loading Unlocking...
Login

Join to view Object Category Recognition and access 3M+ class-specific study document.

or
We will never post anything without your permission.
Don't have an account?
Sign Up

Join to view Object Category Recognition 2 2 and access 3M+ class-specific study document.

or

By creating an account you agree to our Privacy Policy and Terms Of Use

Already a member?