Fig. 3.

An illustration of ESM and ASSM. Firstly, the low resolution feature maps from the stage $S_i$ are resized to the same size $H\times W$ with the input image by using bilinear interpolation up-sampling. Then all high resolution feature maps are reduced to a feature map by using $1\times 1$ convolutions. Finally each pixel value of the obtained feature map is converted to a probability by using Sigmoid function $\sigma (·)$, and the prediction image of the $S_i$ stage is obtained. (a) ESM: the edge supervision is achieved by comparing between the obtained edge prediction image $S_{edge}^i$ and the corresponding edge Ground Truth (GT) $G_{edge}$ based on Eq.(1). (b) ASSM: the auxiliary semantic supervision is achieved by comparing between the obtained coarse segmented image $S_{mask}^i$ and the corresponding Ground Truth (GT) of segmentation mask $G_{mask}$ based on Eq.(2).