Undestanding why USI3D is unsupervised

[chinglamchoi]

Hello, happy Chinese New Year! I'm having trouble understanding why USI3D is unsupervised and would appreciate any help. The functions trainer.dis_update and trainer.gen_update in train.py seem to require passing in ground truth albedo, shading and mask maps, and then does supervised reconstruction losses. Does USI3D work in training for a dataset without these maps being available?

Thank you (2nd Feb, 2022)!