How to calculate multiview point clouds and how to implement "Patch emb"?

As mentioned in the paper "We augment each foreground pixel with an (x; y; z) surface coordinate, giving an image of shape 256 × 256 × 7.", what information does the 7 contain and how to calculate these information with 20 images? And what folder is the "Patch emb" code placed in? thanks for every answer!