Generative Models for Visual Objects and Object Recognition via Bayesian Inference

Posted in Science on September 06, 2008


Generative Models for Visual Objects and Object Recognition via Bayesian Inference

Lecture slides:

  • Generative Models for Visual Objects and Object Recognition via Bayesian Inference
  • Plato said…
  • How many object categories are there?
  • So what does object recognition involve?
  • Verification: is that bus?
  • Detection: are there cars?
  • Identification: is this a picture of Mao?
  • Object categorization
  • Scene and context categorization
  • Challenges 1: view point variation
  • Challenges 2: illumination
  • Challenges 3: occlusion
  • Challenges 4: scale
  • Challenges 5: deformation
  • Challenges 6: background clutter
  • History: single object recognition
  • Challenges 7: intra-class variation
  • History: early object categorization
  • Scenes, Objects, and Parts
  • Object categorization: the statistical viewpoint
  • Discriminative
  • Generative
  • Three main issues
  • Representation
  • Learning
  • Recognition
  • Bag-of-words models
  • Related works
  • Object - bag of words
  • Analogy to documents
  • learning - recognition
  • Representation
  • 1. Feature detection and representation
  • 2. Codewords dictionary formation
  • Image patch examples of codewordsImage codewords
  • 3. Image representation
  • Representation
  • Learning and Recognitio
  • 2 case studies
  • First, some notations
  • Case #1: the Na Naïve ve Bayes model
  • Confusion matrix
  • Case #2: Hierarchical Bayesian text models
  • Case #2: the pLSA model
  • Case #2: Recognition using pLSA
  • Case #2: Learning the pLSA parameters
  • Learning and Recognition
  • Invariance issues
  • Model properties
  • Weakness of the mode
  • part-based models
  • Problem with bag-of-words
  • Overview of section
  • Representation
  • Model: Parts and Structure
  • Representation
  • Example scheme
  • Sparse representation
  • History of Idea
  • The correspondence problem
  • Connectivity of parts
  • Different graph structures
  • Regions or pixels
  • How to model location?
  • Explicit shape model
  • Shape
  • Euclidean & Affine Shape
  • Translation-invariant shape
  • Affine Shape Density
  • Other invariance methods
  • Representation of appearance
  • Representation of occlusion
  • Representation of background clutter
  • Learning
  • Learning situations
  • Learning using EM
  • Example scheme, using EM for maximum likelihood learning
  • Priors
  • Learning Shape & Appearance simultaneously
  • Learn appearance then shape
  • Discriminative training
  • Number of training images
  • Number of training examples
  • Parts and Structure models Summary
  • Recognition
  • What task?
  • Efficient search methods
  • Parts and Structure demo
  • Demo images
  • Demo Web Page
  • How much does shape help?
  • combined segmentation and recognition
  • Aim
  • In this section: brief paper reviews
  • Image parsing: Tu, Zhu and Yuille 2003
  • OBJCUT: shape prior --Layered Pictorial Structures (LPS)
  • OBJCUT
  • OBJCUT: Results
  • LOCUS model
  • Summary
  • List properties of ideal recognition system
  • Online resources

Author: Fei Fei Li, Princeton University

Watch Video

Tags: Science, Lectures, Computer Science, Machine Learning, VideoLectures.Net, Computer Vision, Image Analysis, Bayesian Learning