Perspective Transformer Nets: Learning Single-View 3D Object Reconstruction without 3D Supervision

   Abstract