AI2D stands for AI2 Diagrams. This dataset includes more than 5,000 diagrams on elementary school natural science topics with more than 15,000 rich annotations. It also includes the ground truth syntax analysis of the diagrams and more than 15,000 corresponding multiple-choice questions.

The AI2D dataset establishes four types of icon elements: 'blobs', written text, arrows and arrowheads. In the layout segmentation and parsing diagrams of the AI2D dataset, red represents blobs, blue represents written text, green represents arrows, and orange represents arrowheads.