Could any one help to understand if the ground truth sdf computation for the implicit function learning necessarily need watertight meshes ?