CRAPES:Cross-modal Annotation Projection for Visual Semantic Role Labeling

26431

Knowledge Triples