图像到图像转换（Image-to-Image Translation, I2IT）是在保留图像的内容特征条件下，将源域图像转换成目标域图像的过程，旨在学习不同域图像之间的映射。因I2IT应用广泛，如图像风格迁移、图像语义分割、图像修复和图像超分辨率等，图像转换任务一直是计算机视觉领域中研究的热点和重点，并在近年来因深度学习的蓬勃发展取得了显著进展。其中，基于解耦内容-风格特征表示的无监督模型是图像转换的重要方法。本文从内容表示和风格表示两方面梳理了此类模型的发展历程；总结了图像转换任务中常用的数据集和评价指标，同时比较了经典模型在不同数据集上的效果；最后对基于解耦内容-风格特征表示的无监督图像转换的研究进行总结与展望。

Image-to-Image Translation (I2IT) aims to learn the mapping between images of different domains. Image translation is a hot topic in computer vision since I2IT models can be applied in various fields, including image style transfer, semantic segmentation, and image super-resolution. With the booming development of deep learning, numerous effective I2IT models have emerged in recent years. Among them, unsupervised models based on disentangled content-style representation are essential methods. In this paper, we first review the development of such models in terms of content and style representation. Then, we summarize the common datasets and metrics used in I2IT tasks and compare the results of various models. Finally, we assess and forecast future trends in I2IT development.