This paper studies the context aggregation problem in semantic image
segmentation. The existing researches focus on improving the pixel
representations by aggregating the contextual information within individual
images. Though impressive, these methods neglect the significance of the
r