This paper addresses a challenging problem: classification on unseen viewpoints by training on seen viewpoints. The authors propose a novel model inspired by human binocular vision to addess this problem. Generally speaking, the paper is well organized and motivated. The evaluation part is a bit weak, but the rebuttal provided by the authors has remedy this issue. This work would be an interesting work for the computer vision community. The authors are encouraged to address the issues mentioned by the reviewers.