Unifed Structured Learning For Simultaneous Human Pose Estimation And Garment Attribute Classifcation

In this paper, we utilize structured learning to simultaneously address t-wo intertwined problems: human pose estimation (HPE) and garment attributeclassifcation (GAC), which are valuable for a variety of computer vision andmultimedia applications. Unlike previous works that usually handle the twoproblems separately, our approach aims to produce the optimal joint estimationfor both HPE and GAC via a unifed inference procedure. To this end, we adopta preprocessing step to detect potential human parts from each image (i.e. aset of “candidates”) that allows us to have a manageable input space. In thisway, the simultaneous inference of HPE and GAC is converted to a structuredlearning problem, where the inputs are the collections of candidate ensembles,the outputs are the joint labels of human parts and garment attributes, and thejoint feature representation involves various cues such as pose-specifc features,garment-specifc features, and cross-task features that encode correlations be-tween human parts and garment attributes. Furthermore, we explore the “strongedge” evidence around the potential human parts so as to derive a more powerfulrepresentation for our oriented human part. These evidence can be seamlesslyintegrated into our structured learning model as a kind of energy function. Thelearning process is performed by the structured Support Vector Machines (SVM)algorithm. As the joint structure of the two problems is a cyclic graph, whichhinders an efcient inference, we instead compute the approximate optima via aniterative process, where in each iteration the variables of one problem are fxed,i.e. an inference problem on a tree. In this way, the optimal solutions can be ef-fciently computed by a dynamic programming algorithm. Experimental results demonstrated on two benchmark datasets show the state-of-the-art performanceof our approach.