A Zero-shot Learning Method with a Multi-modal Knowledge Graph

${\tilde{F}}_{s+u}$ represents the result of two-layer network activation

extracted classifier weights for seen-class (denoted with W${}_i$)

CNN is used to extract the visual representations for seen-classes, denoted as F${}_{C_s}$.generative adversarial network (GDAN)，visual representations of all classes F${}_{s+u}$

额外信息

作者未提供代码