| 论文简介 |
Image-to-image translation is a rapid-growing research field in deep learning. It aims to learn the mapping relation between two different domains. Although the existing Generative Adversarial Network(GAN)-based methods have achieved respectable results in this field, there are still some limitations in generating high-quality images for data augmentation. In this work, we focus on image-to-image translation task with the presence of artifacts and the lack of details. To solve these issues, we propose a details fusion generative adversarial network,which consists of details branch,transfer branch, adaptive module and fusion module. By introducing the dual branch design, the proposed model could enhance the generated results with corresponding style and content. Extensive experiments suggest that our model generates more satisfactory images than the competing methods on data augmentation task. |
(创新港)
