Abstract: Image translation aims to achieve conversion between multiple sets of images in different fields, and at the same time needs to constrain the consistency of the distribution of the sample space and the target space. The article aims to find the combination of conditional generative adversarial networks and image translation problems. Firstly, it introduces the characteristics of the datasets , pointed out the difficulty of image translation in different datasets; secondly, derived different ways of algorithm implementation from mathematical expressions, properties and objective function design methods; divided existing image translation into three categories - paired image translation and unpaired image translation and multi-domain image translation, and obtained the image translation categories corresponding to different application scenarios: that is, high-definition tasks correspond to paired image translation, low-cost tasks correspond to unpaired image translation, and diversified tasks correspond to multi-domain image translation; The image quality evaluation method is divided into subjective image quality evaluation and objective image quality evaluation, and the applicable scope of full reference image and no reference image quality evaluation in objective image quality evaluation is analyzed. Finally, we summarized the progress of conditional generative adversarial networks in image translation, and analyzed the algorithm to point out the problems that need to be solved in the future such as mode collapse, model interpretability and few samples.