Visual7W Visual Question Answering Dataset

Visual7W is a dataset for understanding image content. It performs visual question answering tasks by describing image regions in text and their associations. The dataset contains not only the image itself, but also questions and answers related to the content of the image region.

Visual7W is a subset of the Visual Genome dataset, containing 47,300 COCO dataset images, 327,929 question-answer pairs, 1,311,756 human-generated multiple-choice questions, and 561,459 object groundings covering 36,579 categories.

Visual7W's questions are mainly composed of What, Where, How, When, Who, Why, and Which. The questions are multiple-choice, and each question has four candidate answers.