Structure preserving implicit shape encoding via flow regularization

Sven Dummer, Nicola Strisciuglio, Christoph Brune

Keywords: latent space, shape reconstruction, implicit neural representation, neural ODE

Abstract: Recently, implicit neural representations (INRs) emerged as an effective method for reconstructing shapes. Several of such methods transform templates to target shapes. Current template-based methods lack proper regularization. In this work, we add a novel regularization to a deformable template approach and discuss the benefits of this regularization with a simple test case.