revise：In order to effectively carry out multi-task learning we use an end-to-end method combined with IIL to jointly supervise and learn the two branches In particular both the edge feature map and s

To ensure efficient implementation of multi-task learning, we adopt an end-to-end approach that combines IIL to simultaneously supervise and learn the two branches. Specifically, the edge feature map and segmentation feature map are both encoded using binary values.