| task_categories: | |
| - image-text-to-text | |
| tags: | |
| - chart-qa | |
| - visual-grounding | |
| # CCQA: Curriculum Chart Question Answering | |
| **CCQA** (Curriculum Chart Question Answering) is a three-level curriculum dataset introduced in the paper [CURV: Enhancing Chart Understanding Through Curriculum Visual Grounded Reasoning](https://huggingface.co/papers/2608.02833). | |
| - **Paper:** [CURV: Enhancing Chart Understanding Through Curriculum Visual Grounded Reasoning](https://huggingface.co/papers/2608.02833) | |
| - **Project Page:** https://xhguo7.github.io/CURV/ | |
| - **Repository:** https://github.com/xhguo7/CURV | |
| ## Dataset Summary | |
| CCQA is generated entirely from templates, with no model in the loop. Every question, reasoning step, grounding box, and answer is computed deterministically from plotting data, guaranteeing ground-truth correctness at scale. The curriculum systematically progresses from basic single-operation reasoning to complex multi-chart compositional tasks. | |
| ## Citation | |
| ```bibtex | |
| @article{guo2026curv, | |
| title={CURV: Enhancing Chart Understanding Through Curriculum Visual Grounded Reasoning}, | |
| author={Guo, Xuehang and Zhang, Pingyue and Zhang, Ruiyi and Wang, Zhenhailong and Lyu, Hanrui and Ji, Heng and Sun, Tong and Wang, Qingyun and Li, Manling}, | |
| journal={arXiv preprint arXiv:2608.02833}, | |
| year={2026} | |
| } | |
| ``` |