332F/visual_genome

Visual Genome enable to model objects and relationships between objects. They collect dense annotations of objects, attributes, and relationships within each image. Specifically, the dataset contains over 108K images where each image has an average of 35 objects, 26 attributes, and 21 pairwise relationships between objects.