Recognizing materials from virtual examples

Abstract

Due to the strong impact of machine learning methods on visual recognition, performance on many perception task is driven by the availability of sufficient training data. A promising direction which has gained new relevance in recent years is the generation of virtual training examples by means of computer graphics methods in order to provide richer training sets for recognition and detection on real data. Success stories of this paradigm have been mostly reported for the synthesis of shape features and 3D depth maps. Therefore we investigate in this paper if and how appearance descriptors can be transferred from the virtual world to real examples. We study two popular appearance descriptors on the task of material categorization as it is a pure appearance-driven task. Beyond this initial study, we also investigate different approach of combining and adapting virtual and real data in order to bridge the gap between rendered and real-data. Our study is carried out using a new database of virtual materials VIPS that complements the existing KTH-TIPS material database. © 2012 Springer-Verlag.