We propose USTNet, a novel deep learning approach designed for shape-to-shape translation from unpaired domains in an unsupervised manner. The core of our lies disentangled representation that factors out the discriminative features 3D shapes into content and style codes. Given input multiple domains, USTNet disentangles their codes contain distinctive traits across domain-invariant traits. By ...

Computer-aided design of porous structures is a challenging task because of the high degree of irregularity and intricacy associated with the geometries. Most of the existing design approaches either target designing artifacts with regular-shaped pores or reconstructing geometric models from existing porous objects. For regular porous structures, it is difficult to control the pore shapes and d...