BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:MSR-Lecture: Generative Models of Images of Objects - Ali Eslami\,
  University of Edinburgh
DTSTART:20130503T090000Z
DTEND:20130503T100000Z
UID:TALK44765@talks.cam.ac.uk
CONTACT:32739
DESCRIPTION:We first address the question of how to build a 'strong' proba
 bilistic model of object shapes (in the form of binary silhouettes). We de
 fine a 'strong' model as one which meets two requirements: 1. Realism – 
 samples from the model look realistic\, and 2. Generalization – the mode
 l generates samples that differ from training examples. We consider a clas
 s of models known as Deep Boltzmann Machines and show how a strong model o
 f shape can be constructed using a specific form of DBM which we call the 
 'Shape Boltzmann Machine' (ShapeBM).\n \nWe also present a generative fram
 ework for modelling RGB images of objects. Building on the ShapeBM\, our m
 odel employs a factored representation to reason about appearance and shap
 e variability across datasets of images. Parts-based segmentations of obje
 cts are obtained simply by performing probabilistic inference in the propo
 sed model. We apply the model to several challenging datasets which exhibi
 t significant shape and appearance variability\, and find that it obtain
 s results that are comparable to the state-of-the-art.
LOCATION:Small Lecture Theatre\, Microsoft Research Ltd\, 21 Station Road\
 , Cambridge\, CB1 2FB
END:VEVENT
END:VCALENDAR