What AI says about ChartMaster

ChartMaster is a framework for chart-to-code generation that enables multimodal large language models to convert chart images into executable code for automated data analysis, report generation, and intelligent Q&A. It introduces ReChartPrompt, a large-scale dataset of real-world chart–image/code/instruction triplets derived from 30,071 arXiv papers (~240K samples) to improve robustness across diverse chart designs. It also presents ChartSimRL, a reinforcement learning method based on Group Relative Policy Optimization with a chart similarity reward to ensure generated code faithfully matches the original charts both semantically and visually.