How to train the human portrait drawing model?

[Harryoung]

What confused me most is the training schema, i.e., in a segmentation way or a GAN way. In the original work, https://github.com/yiranran/APDrawingGAN, the model is trained in a GAN way. I wonder if the U-2-Net is trained in a segmentation way, i.e., directly predicts the sketch which is consisted of 0/1 pixels. Thanks!