S. M. Eslami, N. Heess, and J. Winn
A good model of object shape is essential in applications such as segmentation, object detection, inpainting and graphics. For example, when performing segmentation, local constraints on the shape can help where the object boundary is noisy or unclear, and global constraints can resolve ambiguities where background clutter looks similar to part of the object. In general, the stronger the model of shape, the more performance is improved. In this paper, we use a type of Deep Boltzmann Machine that we call a Shape Boltzmann Machine (ShapeBM) for the task of modeling binary shape images. We show that the ShapeBM characterizes a strong model of shape, in that samples from the model look realistic and it can generalize to generate samples that differ from training examples. We find that the ShapeBM learns distributions that are qualitatively and quantitatively better than existing models for this task.
In Proc. Conf. Computer Vision and Pattern Recognition (to appear)