Fig. 2From: Multimodal few-shot classification without attribute embeddingPredicted probability of attributes versus ground truthBack to article page