ConKeD: Multiview contrastive descriptor learning for keypoint-based retinal image registration | Publicación