christian-bick/edugraph-exercises

EduGraph Exercises Dataset

EduGraph Exercises is a synthetic ML dataset of math-related visual problems, precisely labeled for training AI models in the education sector.

Every image in this dataset is programmatically generated using the EduGraph Ontology to ensure that visual features are mathematically bound to their pedagogical labels.

Quick Links

• Generation Engine: GitHub Repository (Contribute new generators or views!)
• Ontology: EduGraph Ontology (Semantic definitions for all labels)
• Coverage Explorer: coverage.edugraph.io (Explore standards coverage and generated tasks)

Dataset Description

The dataset follows a Label-Driven Generation paradigm. Instead of labeling existing images, the system uses mathematical constraints to generate problems that satisfy specific educational goals.

Example Modules

The dataset currently contains exercises for:

• Arithmetic: Addition, Subtraction, Multiplication, and Division (Vertical and Box layouts).
• Counting: Object counting with various shapes and symbols.
• Comparison: Greater than, less than, and equality of numbers.
• Time: Analog clock reading at various intervals.

Structure & Metadata

File Layout

The dataset is split into train and validation folders. Each split contains:

• metadata.jsonl: The primary entry point containing labels and image references.
• [module]/: Subdirectories containing the actual PNG images.

Metadata Schema

Each entry in metadata.jsonl contains:

• file_name: Path to the image file (relative to the split root).
• tags: List of semantic labels from the EduGraph Ontology.
• solution: Boolean indicating whether the image shows a solved state.

Versioning & Integrity

The dataset version (e.g., v0.6.0-02) is strictly tied to the version of the EduGraph Ontology it was generated with. This ensures that labels like http://edugraph.io/edu/ArabicNumerals have stable, mathematically consistent meanings across all training samples.

License

This dataset is released under the Apache 2.0 License.