This dataset is designed for research in Deep Learning for Geometry Problem Solving (DL4GPS) and accompanies the survey paper A Survey of Deep Learning for Geometry Problem Solving. It aims to provide a structured resource for evaluating and training AI models, particularly multimodal large language models (MLLMs), on mathematical reasoning tasks involving geometric contexts.

The dataset provides a collection of geometry problems, each consisting of a textual question and a corresponding image.

For a continuously updated reading list of papers on Deep Learning for Geometry Problem Solving, refer to the official GitHub repository.

Data Structure

Each problem instance in the dataset includes the following fields:

• class: The category of the geometry problem.
• id: A unique identifier for each problem.
• question: The textual description of the geometry problem.
• option: Multiple-choice options for the answer, if applicable.
• answer: The correct answer to the geometry problem.
• task_class: A classification of the task involved.
• Attributes: Additional attributes or metadata about the problem.
• image: The image of the geometric diagram associated with the problem.