ReferencesNumerical ExperimentsApplication to images of cluttered objects. ASelection of 14 of the N = 500 data points.BChanges of the parameters Wand Tfor the algorithm with H= 8 hidden units. Each row shows Wand Tfor the specified EM iteration. CFeature vectors at different iterations stages displayed as point in color space. Black circles are the current model values and grey circles those of the previous iterations.Combining Masks and FeaturesAIllustration of how two object masks and features combine to generate an image (without noise).BGraphical model of the generation process with hidden permutation variable .The Generative ModelAbstractWe study unsupervised learning in a probabilistic generative model for occlusion. The problem of occlusion is addressed from the perspective of multiple-causes models such as NMF , sparse coding , or ICA.The model features- binary hidden variables encoding object presence- a vectorial hidden variable for object proximities
This is the end of the preview.
access the rest of the document.