Despite the success of InfoGAN’s latent codes in class-varying and attribute-editing in synthetic aperture radar (SAR) images, there lacks enough interpretation of the relation between classes/attributes and latent codes. In this work, we show that such a relation can be described as an approximate Gaussian function. By introducing the rotation angle estimators, we can quantify this relation. Various qualitative and quantitative results reveal that the rotation (view angle) can be calculated by a set of certain latent codes, inversely, the satisfying latent codes can also be estimated when a desired viewing angle is given. In this case, both classes and viewing angles of generated SAR images are completely under users’ control.