Good Recognition is Non-Metric
This addresses a foundational issue in computer vision by challenging assumptions about recognition, potentially influencing algorithm design for broader applications.
The paper tackles the problem of formalizing visual recognition beyond pair matching, showing through meta-analysis and experiments that good recognition algorithms often violate metric properties, and argues for leveraging outside information.
Recognition is the fundamental task of visual cognition, yet how to formalize the general recognition problem for computer vision remains an open issue. The problem is sometimes reduced to the simplest case of recognizing matching pairs, often structured to allow for metric constraints. However, visual recognition is broader than just pair matching -- especially when we consider multi-class training data and large sets of features in a learning context. What we learn and how we learn it has important implications for effective algorithms. In this paper, we reconsider the assumption of recognition as a pair matching test, and introduce a new formal definition that captures the broader context of the problem. Through a meta-analysis and an experimental assessment of the top algorithms on popular data sets, we gain a sense of how often metric properties are violated by good recognition algorithms. By studying these violations, useful insights come to light: we make the case that locally metric algorithms should leverage outside information to solve the general recognition problem.