DeepMend: Learning Occupancy Functions to Represent Shape for Repair

We present DeepMend, a novel approach to reconstruct resto- rations fractured shapes using learned occupancy functions. Existing shape repair approaches predict low-resolution voxelized restorations or smooth restorations, require symmetries access pre-existing complete oracle. represent the of as conjunction an underlying and break surface, which we model functions latent codes neural networks. Given samples from shape, estimate inference loss augmented with penalties avoid empty voluminous restorations. use estimated restoration shape. show results simulated fractures on synthetic real-world scanned objects, real mugs. Compared existing two baseline methods, our work shows state-of-the-art in accuracy avoiding artifacts over non-fracture regions