I have two sets of images with incompletely similar content. I want to compare their differences in texture (style), similar to the CycleGAN loss function (while D aims to distinguish between translated samples G(x) and real samples).