Geometry3k
Multimodal dataset combining images and textual geometric problems, with associated answers, ideal for fine-tuning AI models to solve visual problems.
Approximately 3,000 examples, PNG images and texts, 59.3 MB in Parquet format
MIT
Description
Geometry3k offers 3,002 examples of geometric problems illustrated by an image accompanied by a textual statement and a numerical answer. Each example contains a diagram image, the text of the problem, and the correct solution, making it easy to train multimodal models that can analyze and solve visual mathematical questions.
What is this dataset for?
- Train multimodal models to solve geometric problems
- Develop educational AI assistants specialized in mathematics
- Test the joint understanding of images and text by LLM models
Can it be enriched or improved?
This dataset can be supplemented with new annotations, such as detailed resolution steps, or extended with other types of illustrated math problems. Human work to enrich statements and diagrams would also improve its quality.
🔎 In summary
🧠 Recommended for
- Educational AI researchers
- Multimodal model developers
- Math teachers and students
🔧 Compatible tools
- Hugging Face Datasets
- PIL
- PyTorch
- TensorFlow
💡 Tip
Combine this dataset with other math games to improve the diversity of the problems dealt with.
Frequently Asked Questions
What is the data structure in Geometry3k?
Each entry contains a diagram image, a textual statement of the problem, and an associated numerical answer.
Is this dataset suitable for training multimodal models?
Yes, it is specifically designed for that, by combining images and mathematical text.
What license governs the use of this dataset?
The MIT license allows free use, including commercial use, subject to compliance with the conditions of the license.
