Mesh Deformation Scheme for High Quality 3D Model Reconstruction

This paper proposes a new mesh deformation scheme in a 3D reconstruction in which the performance deteriorates when an object of an unlearned category is input in a 3D model generation network. The proposed method uses a new loss function to generate appropriate 3D reconstruction outputs even for inputs of unlearned categories of 3D models. We compare the performance with the existing method so as to prove the validity of this method. The simulation results show that the proposed loss function improves the generalization ability of the 3D reconstruction with good performance even for inputs of unlearned categories.