FCGNet: Foreground and Class Guided Network for human parsing

Jaehyuk Jang, Yooseung Wang, Changick Kim

Abstract: Highlights•Background and indistinguishable classes make it hard for human parsing.•Spatial foreground information guides to filter out the background.•Existing class information guides to represent indistinguishable classes distinctly.•Graph convolutional networks understand the structural characteristics of the human.•Guiding to predict only existent classes in the image improves the performance.