Retinal vessels segmentation is well known problem in image processing on the medical field. Good segmentation may help doctors take better decisions while diagnose eyes disuses. This paper describes our work taking up the DRIVE challenge which include segmentation on retinal vessels. We invented a new method which combine using of StyleGAN2 and SA-Unet. Our innovation can help any small data set segmentation problem.