On a connection between object localization with a generative template of features and pose-space prediction methods

We address the task of localizing objects from a given object class in an image. The image is represented as a collection of “visual words” at interest points. The generative template of features (GTF) model defines a distribution over visual words and their spatial locations for each part of the object (Sudderth et al., 2005; Fergus et al., 2005). We show how to derive pose-space prediction methods (such as the Hough transform) from the GTF.