Compositional Models

This talk gives an overview of recent work which addresses different computer vision tasks. It describes a research strategy based on compositional models. It describes: (i) low-level representations such as image patches, (ii) segmentation methods to detect candidate regions for objects and object parts, (ii) methods for detecting and parsing humans, and (iv) techniques for estimating 3D shape (e.g., humans) from single images.