[Machine Learning][inrial lecture] and Category Representation

from
http://lear.inrialpes.fr/~verbeek/mlcr.slides.11.12/lecture1verbeek.pdf

• Visual recognition - Objectives

  • Image classification: assigning label to the image
  • Object localization: define the location and the category

• dis of k-means clustering

  • Solution depends heavily on initialization
  • Assignment of data points to clusters is
    only based on the distance to the cluster center
    
    • No representation of the shape of the cluster
    • can be fixed by using simple elliptical shapes