Frontal View Synthesis Based on a Novel GAN with Global and Local Discriminators
Yilong Lu, Tanfeng Sun, Xinghao Jiang, Ke Xu, Bo Zhu · 2019
Although researchers have achieved a relatively high accuracy in face recognition, pose variations are still a big problem in the real life. This paper proposes a novel Generative Adversarial Network with global and local discriminators for frontal view synthesis. Two discriminators are designed to deal with global features and local pixels respectively. This paper also introduces a combined loss function with L1 loss, blur loss, and identity preserving loss. The method is effective for facial analysis tasks with synthesized faces. Experiments show the method can preserve both identities and image quality.