Human Parsing with Edge Enhancement

Human parsing, which aims at segmenting human images according to body parts, attracts many researchers' attention because of its various application perspectives. The problem requires accurate object edge detection on segmentation masks. However, the edge information has not been fully explored in the existing methods for this issue. This paper focuses on enhancing body's edge segmentation performance of the human parsing problem. The proposed enhancement method includes three aspects, which are feature extraction, context information, and edge refinement. To this end, we introduce an edge enhancement module that focuses on edge enhancement and refines the object edges from a coarse predicted segmentation mask. The edge enhancement module, together with a context embedding module and a feature extraction module, is integrated into a state-of-the-art human parsing framework. Our proposed method can significantly improve the performance in edge estimation. The experimental results on LIP dataset and PASCAL-Person-Part dataset validate the excellent performance of our method.