Contrastive Visual and Language Learning for Visual Relationship Detection

28384

Knowledge Triples