Mathematical Image Editing Trajectories for Multimodal Models

MathCanvas-Edit contains 5.2 million step-by-step editing trajectories for mathematical images. The dataset was created by author shiwk24 and was last updated on the Hugging Face platform in November 2025. It forms a core component of the MathCanvas framework for training large multimodal models.

Use Cases

• Training models to predict the next editing step in a mathematical diagram sequence using trajectory data.
• Fine-tuning image-to-image models on the visual editing steps documented in the trajectories.
• Developing models for visual chain-of-thought reasoning by analyzing the step-by-step sequences.
• Benchmarking multimodal models on tasks involving geometry diagram understanding and modification.

Strengths

• Contains 5.2 million data points, providing a large-scale resource.
• Specifically designed for step-by-step editing, a detailed task for multimodal models.

Limitations

• Specific column names, data formats, and sample structure are not provided.
• The geographic and temporal coverage of the source mathematical problems is unknown.
• Potential class imbalance or label noise in the editing trajectories cannot be assessed without sample data.

Provenance

Source

shiwk24 on Hugging Face.

Collection Method

Part of the MathCanvas framework; specific collection method not detailed.

Freshness

Last updated on the platform in November 2025.