Weakly Supervised Annotations for Multi-modal Greeting Cards Dataset