基于深度学习的图像生成与风格迁移技术研究

Research on image generation and style transfer technology based on deep learning

为了解决图像生成期间的问题(如显著区域内容扭曲、目标丢失等),文章提出加入显著性检测网络、设计显著性损失等策略,通过不断迭代优化网络,生成的图像在内容、风格和显著性方面都能达到较好的效果。

In order to solve the problems during image generation,such as distortion of salient region content and loss of targets,the article proposes strategies such as adding a saliency detection network and designing saliency loss.Through continuous iterative optimization of the network,the generated images can achieve good results in terms of content,style,and saliency.