CLOSURE

The CLOSURE dataset consists of 25,200 questions with identical vocabulary but different structure than CLEVR, asked on the same set of images.

BibTex: