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