Oriented and Directional Chamfer Distance Losses for 3D Object Reconstruction from a Single Image

The application of deep learning in the field 3D reconstruction has greatly improved quality object reconstruction. For methods that take point cloud as supervision information, previous research mainly focused on network architecture while setting Chamfer Distance (CD) loss default function. However, CD only contains distance information ignoring directional information. In this paper, we introduce novel losses considering directions can be used a network. These consider both direction and have two specific variants, Oriented (OCD) Directional (DCD). Numerous experiments conducted deformable patch reconstruction, show some classic neural networks for with OCD or DCD achieve better results than those loss.