DrapeNet: Garment Generation and Self-Supervised Draping

Drape unseen garments Recover 3D models from… …images …3D scans initial longer sleeves shorter opened longer trousers Edit garments Figure 1. Our network can drape garments over bodies of different shapes in various poses. To minimize the required amount of supervision, our draping network is trained with physics-based self-supervision and generalizes to multiple garments by being conditioned on latent codes. These can be manipulated to edit specific features of the corresponding garments. Being fully differentiable, our pipeline makes it possible to recover 3D models of garments and bodies from observations such as images and 3D scans.