ChartMimic-500 is a benchmark task that evaluates the ability of multimodal language models to generate code implementations from visual designs. We have conducted a comprehensive study and benchmarking of this task and provide a curated dataset of 500 real-world charts for testing.
This repository provides a detailed overview of the chart to code task, automatic evaluation metrics, and a collection of multimodal prompting methods.
To get started with ChartMimic-500, please refer to the instructions in the repository. We provide a detailed guide on how to use the benchmark, apply the evaluation metrics, and fine-tune the models.
We welcome contributions to ChartMimic-500! If you have suggestions for improvements or additions, please open an issue or submit a pull request.