We study the problem of few-shot open-set recognition (FSOR), which learns a
recognition system capable of both fast adaptation to new classes with limited
labeled examples and rejection of unknown negative samples. Traditional
large-scale open-set methods have been shown ineffective f