Ai2D: 1,000-10,000 Scientific Diagrams for Visual Reasoning

Ai2D contains between 1,000 and 10,000 scientific diagrams with corresponding text annotations, published by Aniruddha Kembhavi and the Allen Institute for AI in 2016. The dataset is designed to support research in diagrammatic reasoning and visual question answering within the scientific domain.

Use Cases

• Visual Question Answering (VQA) using diagram images and text labels
• Object detection for identifying diagram components like arrows and text boxes
• Structural parsing to map relationships between visual elements and text

Strengths

• 1,000-10,000 annotated records
• Peer-reviewed methodology from arXiv:1603.07396
• Multimodal support for both image and text processing

Limitations

• Small sample size relative to modern vision-language datasets
• Domain-specific focus restricted to K-12 level scientific diagrams

Provenance

Source

Allen Institute for AI (Kembhavi et al., 2016)

Collection Method

annotated

Time Range

2016

Freshness

Last updated March 2024; original data published in 2016.