We have seen remarkable recent progress in computational visual recognition, producing systems that can classify objects into thousands of different categories with increasing accuracy. However, one question that has received relatively less attention is "what labels should recognition systems output?" This paper looks at the problem of predicting category labels that mimic how human observers would name objects. This goal is related to the concept of entry-level categories first introduced by psychologists in the 1970s and 1980s. We extend these seminal ideas to study human naming at large scale and to learn computational models for predicting entry-level categories. Practical applications of this work include improving human-focused computer vision applications such as automatically generating a natural language description for an image or text-based image search.
Computational visual recognition is beginning to work. Although far from solved, algorithms for analyzing images have now advanced to the point where they can recognize or localize thousands of object categories with reasonable accuracy.3, 14, 24, 25 While we could predict any one of many relevant labels for an object, the question of "What should I actually call it?" is becoming important for large-scale visual recognition. For instance, if a classifier were lucky enough to get the example in Figure 1 correct, it might output Cygnus Colombianus, while most people would probably simply say swan. Our goal is to learn models to map from specific, encyclopedic terms (Cygnus Colombianus) to how people might refer to a given object (swan).