deformable image registration is a fundamental task in medical image
analysis, aiming to establish a dense and non-linear correspondence between a
pair of images. Previous deep-learning studies usually employ supervised neural
networks to directly learn the spatial transformation from