Multiple Instance Classification or: How I Learned to Evaluate Local Image Descriptors

Multiple instance classification (MIC) is a kind of supervised learning, where data are represented as bags and each bag contains many instances. Training bags are given a label and the system tries to learn how to label bags, without necessarily learning how to label each instance individually. In this paper, we apply concepts drawn from MIC to the realm of content-based image retrieval, where images are described as bags of visual local descriptors. In particular, we purport the use of classifiers, following the different MIC paradigms, to evaluate the effectiveness of any local descriptor.