Fetching the paper…

COVR: A test-bed for Visually Grounded Compositional Generalization with real images · Around