Keywords to Visual Categories: Multiple-Instance Learning Forweakly Supervised Object Categorization

Abstract

Conventional supervised methods for image categoriza- tion rely on manually annotated(labeled) examples to learn good object models, which means their generality and scal- abilitydependsheavilyon the amountof humaneffort avail- able to help train them. We propose an unsupervised ap- proach to construct discriminative models for categories specified simply by their names. We show that multiple- instance learning enables the recovery of robust category models from images returned by keyword-based search en- gines. By incorporatingconstraints that reflect the expected sparsity of true positive examples into a large-margin ob- jective function, our approach remains accurate even when the available text annotations are imperfect and ambigu- ous. In addition, we show how to iteratively improve the learned classifier by automatically refining the representa- tion of the ambiguously labeled examples. We demonstrate our method with benchmark datasets, and show that it per- forms well relative to both state-of-the-artunsupervised ap- proaches and traditional fully supervised techniques.