We propose an approach to the problem of detecting and segmenting generic object classes that combines three "off the shelf" components in a novel way. The components are a generic image segmenter that returns a set of "super pixels" at different scales; a generic classifier that can determine if an image region (such as one or more super pixels) contains (part of) the foreground object or not; and a generic belief propagation (BP) procedure for tree-structured graphical models. Our system combines the regions together into a hierarchical, tree-structured conditional random field, applies the classifier to each node (region), and fuses all the information together using belief propagation. Since our classifiers only rely on color and texture, they can handle deformable (non-rigid) objects such as animals, even under severe occlusion and rotation. We demonstrate good results for detecting and segmenting cows, cats and cars on the very challenging Pascal VOC dataset.