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

Abstract

Many methods have been proposed to solve the image classification problem for a large number of categories. Among them, methods based on tree-based representations achieve good trade-off between accuracy and test time efficiency. While focusing on learning a tree-shaped hierarchy and the corresponding set of classifiers, most of them [11,2,14] use a greedy prediction algorithm for test time efficiency. We argue that the dramatic decrease in accuracy at high efficiency is caused by the specific design choice of the learning and greedy prediction algorithms. In this work, we propose a classifier which achieves a better trade-off between efficiency and accuracy with a given tree-shaped hierarchy. First, we convert the classification problem as finding the best path in the hierarchy, and a novel branchand-bound-like algorithm is introduced to efficiently search for the best path. Second, we jointly train the classifiers using a novel Structured SVM (SSVM) formulation with additional bound constraints. As a result, our method achieves a significant 4.65%, 5.43%, and 4.07% (relative 24.82%, 41.64%, and 109.79%) improvement in accuracy at high efficiency compared to state-of-the-art greedy "tree-based" methods [14] on , SUN [32] and ImageNet 1K [9] dataset, respectively. Finally, we show that our branch-and-bound-like algorithm naturally ranks the paths in the hierarchy (Fig. 8) so that users can further process them.

Joint Hierarchical Category Structure Learning and Large-Scale Image Classification

“…For instance, Li et al [1] constructed ImageNet according to WordNet, a semantic hierarchy taxonomy unrelated to visual effects. Although WordNet has been widely applied to image classification [53], the visual attributes are always ignored.…”

“…In contract, our method provides a distribution of hierarchical memberships for image categories based on spectral clustering, in which the best path algorithm is developed to avoid error propagation. The closest related work is [53], where the best path is learned by the structured SVM, leading to high computational complexity. We make classification predictions based on the best path algorithm depending on the hierarchical structure.…”

Learning mid-level features from object hierarchy for image classification

“…To achieve better trade-off between efficiency and accuracy with the object hierarchy, Sun et al [25] proposed to use the branch-and-bound technique for efficient classification. The test image in the standard tree-based algorithm traverses only single path in the tree from the root to a leaf based on the linear classifiers associated with every internal node.…”

“…This implies that if a mistake occurs at one node, it will be propagated to a leaf node and cannot be recovered. Sun et al [25] overcame this limitation by exploring more than one path in the tree-shaped hierarchy and typically finish in sublinear time.…”

Object Classification Using a Semantic Hierarchy

“…Gao et al [8] further improved the method by allowing overlapping object classes at different child nodes. Sun et al [9] proposed to use the branch-and-bound technique on object hierarchy for efficient classification. Object hierarchy has also been used to improve image retrieval [10] and to provide accuracy-specificity trade-offs in large scale recognition [11].…”