ET-ImageStyler: Efficient Text-Guided Image Style Transfer

One of the main topics of current research is image style transfer. The level of style transfer has steadily increased with the introduction of new models. However, most of the research focuses on image-driven style transfer. In other words, these models require an image of the target style to complete the transfer. In most cases, it is difficult to find a precise style image. Therefore, text-guided image style transfer is important and convenient. However, existing lightweight text-based image style transfer models suffer from the problem of limited feature quality expression. Therefore, this paper proposes ET-ImageStyler. This is an efficient text-guided image style transfer method that incorporates an attention mechanism. In this study, an effective multi-scale attention module is introduced, and it is imaginatively combined with conventional text-based style transfer. This model can complete style transfer with a few parameters. The ability of the image produced by ET-ImageStyler to better maintain the structural information of the original content image has been verified through rigorous trials. In addition, the image has better feature quality than traditional methods.