This is a visual question answering dataset, where the questions and answers are automatically generated from image descriptions in MS-COCO. This dataset can be used to solve problems such as image-based question answering.

The dataset includes:

• 123,287 images
• 78,736 training questions
• 38,948 test questions
• Four question types: object, number, color, position
• The answer is described in one word only.