MIT and IBM Propose ChartNet, the Largest Synthetic Chart Dataset to Date, Generating 1.5 Million Diverse Chart Samples

▤ Summary

MIT, the MIT-IBM Watson AI Lab, and IBM Research introduced ChartNet, a 1.5 million-sample synthetic chart dataset for chart understanding. The dataset combines chart images, plotting code, tabular data, descriptions, QA pairs, and grounding annotations to support multimodal training. It also includes real-world charts, manually annotated samples, and security-focused data to broaden model capabilities. Experiments show that models fine-tuned on ChartNet outperform larger baselines and even GPT-4o on several chart tasks. The release positions ChartNet as a new foundation for chart reconstruction, data extraction, and chart summarization research.

▦ Linked Companies