SBS Figures: Pre-training Figure QA from Stage-by-Stage Synthesized Images

TL;DR

SBSFigures is a new dataset created through a stage-by-stage pipeline that enables efficient pre-training of figure QA models with fully annotated, diverse synthetic figures, reducing manual effort and improving training effectiveness.

Contribution

The paper introduces SBSFigures, a novel synthetic figure dataset generated without manual annotation, enhancing pre-training for figure question answering tasks.

Findings

Pre-training with SBSFigures improves QA performance on real-world data.

The pipeline reduces code errors and increases diversity in synthetic figures.

Efficient training with limited real data is achievable using the pre-trained weights.

Abstract

Building a large-scale figure QA dataset requires a considerable amount of work, from gathering and selecting figures to extracting attributes like text, numbers, and colors, and generating QAs. Although recent developments in LLMs have led to efforts to synthesize figures, most of these focus primarily on QA generation. Additionally, creating figures directly using LLMs often encounters issues such as code errors, similar-looking figures, and repetitive content in figures. To address this issue, we present SBSFigures (Stage-by-Stage Synthetic Figures), a dataset for pre-training figure QA. Our proposed pipeline enables the creation of chart figures with complete annotations of the visualized data and dense QA annotations without any manual annotation process. Our stage-by-stage pipeline makes it possible to create diverse topic and appearance figures efficiently while minimizing code errors. Our SBSFigures demonstrate a strong pre-training effect, making it possible to achieve efficient training with a limited amount of real-world chart data starting from our pre-trained weights.