The ShadowGAN interprets mountains in everything it sees, even the human form.
The ShadowGAN recognizes the silhouettes of people and then fills them in with generated pictures of mountains, for example. In one of the examples, the AI was trained with countless images of mountains and mountain landscapes, so it is only familiar with natural landscapes and mountains. So it interprets mountains in everything it sees, even the human form. A conditional generative adversarial network (cGAN) is used in the background.