Summary: This paper proposes an end-to-end network which can simultaneously generates and manipulates the face images with desired attributes. The proposed network consists of four parts: generator, discriminator, classifier and connection network. Loss functions are derived for the network.
1. This paper proposes a novel network. Though all component of the network are employed from other work, the structure of the network is novel.
2. This paper provides plenty of experimental results for evaluation. It compares with several recent publications.
3. This paper describes the proposed method and experiments clearly, which makes it easy for other researchers to follow up.
1. The novelty of this paper is limited, since it just propose a new structure of the network. The network is based on traditional GAN, a attribute classifier and a connection network.
2.The explanation of Fig.1 is needed to be supplemented. ex, there is no annotation for ‘yb’, the inconsistency of ‘CN'(in figure) and Cn (in text).
This paper proposes a end-to-end network which can simultaneously generates and manipulates the face images with desired attributes. According to the reported experimental results, it outperforms recently proposed methods. The proposed network either benefits attributes editing or improves the image quality. Though the novelty of this work is limited, it makes GAN network more flexible.