Constructing Category Hierarchies for Visual Recognition

Abstract

Abstract. Class hierarchies are commonly used to reduce the complexity of the classification problem. This is crucial when dealing with a large number of categories. In this work, we evaluate class hierarchies currently constructed for visual recognition. We show that top-down as well as bottom-up approaches, which are commonly used to automatically construct hierarchies, incorporate assumptions about the separability of classes. Those assumptions do not hold for visual recognition of a large number of object categories. We therefore propose a modification which is appropriate for most top-down approaches. It allows to construct class hierarchies that postpone decisions in the presence of uncertainty and thus provide higher recognition accuracy. We also compare our method to a one-against-all approach and show how to control the speed-foraccuracy trade-off with our method. For the experimental evaluation, we use the Caltech-256 visual object classes dataset and compare to stateof-the-art methods.

Learning hierarchical similarity metrics

“…With a few notable exceptions, most works in the literature either focus on developing good classifiers for taxonomy data without learning a metric [4,3,15,16,10,7], or focus on developing good metrics without exploiting the taxonomy structure [22,8,12,9,1,18]. For instance, Hierarchical SVM [4]-an example of the first kind-adapts the basic SVM classifier to learn a linear hyperplane for each category by accumulating contributions from each node along the path from the root to a leaf, but does not yield a metric.…”

“…These local metrics are combined together (Q t ) along the paths of the taxonomy for effective object categorization. cess to this taxonomy structure has been shown to benefit both the accuracy as well as the scalability of learning algorithms [15,16,10].…”

Can Computer Vision Problems Benefit from Structured Hierarchical Classification?

“…About the first condition, high-level hierarchical feature representations can be obtained through biologically inspired design [20,26,33] or example-driven discovery which includes information transfer [9,13] and hierarchy learning [10,16,18,19,29]. In our real-world problems (Sect.…”

“…The motivation is to reduce the complexity of visual recognition problems that have a very large number of instances. To build useful taxonomies, the proposed methods exploit either purely the semantic tag labels [19,29], or purely the visual information [10,18], or both as in [16], where the authors propose a way to learn a "semantivisual" hierarchy that is both semantically meaningful and close to the visual content.…”

Find the Best Path: An Efficient and Accurate Classifier for Image Hierarchies

“…LLC feature with 21K dimension is used in order to recreate the one-versus-all accuracy reported in [14]. The relaxed hierarchy method [14] is a very strong baseline since it outperforms many other "tree-based" method [24,16]. Therefore, five hierarchies corresponding to relaxed degree ρ = {0.5, 0.6, 0.7, 0.8, 0.9} (the bigger value the less relax) are trained for comparison.…”