AutoGeo: Automating Geometric Image Dataset Creation for Enhanced Geometry Understanding
Project Overview
The document introduces AutoGeo, an innovative automated system developed to generate extensive geometric image datasets, thereby filling a significant gap in the availability of high-quality resources for mathematical reasoning in education. By creating the AutoGeo-100k dataset, which includes 100,000 meticulously curated image-text pairs centered on geometry, the system significantly improves the capabilities of multimodal large language models (MLLMs) in tasks related to geometric understanding, such as captioning and question-answering. The research underscores the challenges associated with generating effective geometric datasets and outlines a novel pipeline that enhances the efficiency of the data creation process, ensuring both validity and diversity while simultaneously reducing associated costs. This advancement not only aids in the enhancement of educational tools but also contributes to the broader goals of integrating AI into educational settings to improve learning outcomes in mathematical reasoning.
Key Applications
AutoGeo for generating geometric image datasets
Context: Educational and research contexts focused on geometry for students and researchers in mathematics
Implementation: AutoGeo utilizes an augmented geometry clause system and a rule-based clause selector to automatically generate geometric images and text descriptions
Outcomes: Creation of AutoGeo-100k, a dataset of 100k high-quality geometry image-text pairs, leading to improved performance in geometric tasks for MLLMs
Challenges: Initial limitations in the availability of high-quality geometric datasets hindered the development of AI tools in mathematical reasoning
Implementation Barriers
Data scarcity
Existing geometric datasets are limited in size and quality, which restricts the effectiveness of MLLMs in understanding geometric concepts.
Proposed Solutions: AutoGeo addresses this by automating the generation of geometric image datasets, providing a scalable solution.
Complexity in geometric reasoning
Geometric reasoning involves intricate relationships and requirements for precise logical generation, which current models struggle to handle.
Proposed Solutions: The introduction of a comprehensive geometric definition system and structured clause generation in AutoGeo mitigates these complexities.
Project Team
Zihan Huang
Researcher
Tao Wu
Researcher
Wang Lin
Researcher
Shengyu Zhang
Researcher
Jingyuan Chen
Researcher
Fei Wu
Researcher
Contact Information
For information about the paper, please contact the authors.
Authors: Zihan Huang, Tao Wu, Wang Lin, Shengyu Zhang, Jingyuan Chen, Fei Wu
Source Publication: View Original PaperLink opens in a new window
Project Contact: Dr. Jianhua Yang
LLM Model Version: gpt-4o-mini-2024-07-18
Analysis Provider: Openai