Transfer Learning of Object Classes: From Cartoons to Photographs

We consider the important challenge of recognizing a variety of deformable objects in images. Of fundamental importance and particular diculty in this setting is the problem of "outlining" an object, rather than simply deciding on its presence or absence. A major obstacle in learning a model that will allow us to address this task is the need for hand-segmented training images. In this paper we present a transfer learning approach that circumvents this problem by transferring the "essence" of an object from cartoon images to natural images, using a landmark-based model. The use of transfer to create an automatic model-learning pipeline greatly increases our eciency and flexibility in learning novel objects with minimal user supervision. We show that our method is able to automatically learn, detect and localize a variety of classes.