A Ranking Part Model for Object Detection

Abstract

AbstractObject detection has long been considered a binary-classification problem, but this formulation ignores the relationship between examples. Deformable part models, which achieve great success in object detction, have the same problem. We use learning to rank methods to train better deformable part models, and formulates the optimization problem as a generalized convex concave problem. Experiments show that, using same features and similar part configurations, performance of detection by the ranking model outperforms original deformable part models on both INRIA pedestrians and Pascal VOC benchmarks.KeywordsObject DetectionDeformable Part ModelLearning to Rank