Visually grounded few-shot word learning in low-resource settings

Leanne Nortje, Dan Oneata, Herman Kamper

We propose a visually grounded speech model that learns new words and their visual depictions from just a few word-image example pairs. Given a set of test images and a spoken quer…