VisualSem

VisualSem is a knowledge graph designed and curated to support research in vision and language. It is built using BabelNet v4.0 and ImageNet as a starting point and it contains over 101k nodes, 1.9M tuples, and 1.5M glosses and 1.5M images associated to nodes. It is described in detail in our resource paper. In a nutshell, VisualSem includes: 101,244 nodes which are linked to BabelNet ids, and therefore linkable to Wikipedia article ids, WordNet ids, etc (through BabelNet). 13 visually relevant relation types: is-a, has-part, related-to, used-for, used-by, subject-of, receives-action, made-of, has-property, gloss-related, synonym, part-of, and located-at. 1.9M tuples, where each tuple consists of a pair of nodes connected by a relation type. 1.5M glosses linked to nodes which are available in up to 14 different languages. 1.5M images associated to nodes.