COVR: A Test-Bed for Visually Grounded Compositional Generalization with Real Images.

Ben Bogin, Shivanshu Gupta, Matt Gardner, Jonathan Berant

Browse the full EMNLP paper archive.