ChartCap: Mitigating Hallucination of Dense Chart Captioning
Junyoung Lim, Jaewoo Ahn, Gunhee Kim
2025-08-06
Summary
This paper talks about ChartCap, a large dataset with detailed and specific captions for different types of real-world charts to help AI models better understand and describe charts.
What's the problem?
The problem is that vision language models often make mistakes or imagine false details (hallucinations) when describing complex charts, leading to inaccurate or misleading captions.
What's the solution?
ChartCap improves this by providing lots of accurate, type-specific captions that train models to recognize and describe charts more precisely, reducing the chances of hallucination in the captions.
Why it matters?
This matters because better chart descriptions help people understand data more clearly and accurately, making AI tools more reliable for tasks like data analysis and reporting.
Abstract
ChartCap, a large-scale dataset with dense, type-specific captions for real-world charts, improves caption accuracy and reduces hallucinations in vision language models.