Abstract

The learning of hierarchical representations for image classification has experienced an impressive series of successes due in part to the availability of largescale labeled data for training. On the other hand, the trained classifiers have traditionally been evaluated on a handful of test images, which are deemed to be extremely sparsely distributed in the space of all natural images. It is thus questionable whether recent performance improvements on the excessively re-used test sets generalize to real-world natural images with much richer content variations. In addition, studies on adversarial learning show that it is effortless to construct adversarial examples that fool nearly all image classifiers, adding more complications to relative performance comparison of existing models. This work presents an efficient framework for comparing image classifiers, which we name the MAximum Discrepancy (MAD) competition. Rather than comparing image classifiers on fixed test sets, we adaptively sample a test set from an arbitrarily large corpus of unlabeled images so as to maximize the discrepancies between the classifiers, measured by the distance over WordNet hierarchy. Human labeling on the resulting small and model-dependent image sets reveals the relative performance of the competing classifiers and provides useful insights on potential ways to improve them. We report the MAD competition results of eleven ImageNet classifiers while noting that the framework is readily extensible and cost-effective to add future classifiers into the competition.