Human parsing is a fundamental task to estimate semantic parts in a human image such as face, arm, leg, hat, and dress. Recent deep-learning based methods have achieved significant improvements, but collecting training ...