The Natural Language for Visual Reasoning corpora are two language grounding datasets containing natural language sentences grounded in images. The task is to determine whether a sentence is true about a visual input. The data was collected through crowdsourcings, and solving the task requires reasoning about sets of objects, comparisons, and spatial relations. This includes two corpora: NLVR, with synthetically generated images, and NLVR2, which includes natural photographs.
