eduzhai > Applied Sciences > Engineering >

Semantic Graph-enhanced Visual Network for Zero-shot Learning

  • Save

... pages left unread,continue reading

Document pages: 19 pages

Abstract: Zero-shot learning uses semantic attributes to connect the search space ofunseen objects. In recent years, although the deep convolutional network bringspowerful visual modeling capabilities to the ZSL task, its visual features havesevere pattern inertia and lack of representation of semantic relationships,which leads to severe bias and ambiguity. In response to this, we propose theGraph-based Visual-Semantic Entanglement Network to conduct graph modeling ofvisual features, which is mapped to semantic attributes by using a knowledgegraph, it contains several novel designs: 1. it establishes a multi-pathentangled network with the convolutional neural network (CNN) and the graphconvolutional network (GCN), which input the visual features from CNN to GCN tomodel the implicit semantic relations, then GCN feedback the graph modeledinformation to CNN features; 2. it uses attribute word vectors as the targetfor the graph semantic modeling of GCN, which forms a self-consistentregression for graph modeling and supervise GCN to learn more personalizedattribute relations; 3. it fuses and supplements the hierarchicalvisual-semantic features refined by graph modeling into visual embedding. Bypromoting the semantic linkage modeling of visual features, our methodoutperforms state-of-the-art approaches on multiple representative ZSLdatasets: AwA2, CUB, and SUN.

Please select stars to rate!


0 comments Sign in to leave a comment.

    Data loading, please wait...