You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Rethinking Comprehensive Benchmark for Chart Understanding: A Perspective from Scientific Literature

GitHub | Paper

Dataset Details

SCI-CQA introduces a novel evaluation framework inspired by human exams, encompassing 5,629 carefully curated questions, both objective and open-ended.

SCI-CQA consists of a total of 7 files.

The chart directory under chart_test_images.zip contains 2,894 images, which include all the images used in the question-answer pairs. To make it easier to distinguish, the images corresponding to different questions are placed in separate folders within this directory.

The remaining 6 files correspond to different question-answer pairs. We have categorized all the questions into two major types based on chart types: flowcharts (such as model structure diagrams, algorithm flowcharts, etc.) and data charts (such as line charts, bar charts, scatter plots, etc.). All questions are further divided into two categories: perception and reasoning, to facilitate fine-grained evaluation.

License And Copyright

SCI-CQA are licensed under CC BY-SA 4.0. The copyright of all charts belongs to their original authors.

We thank all the chart creators; your excellent charts have helped us build SCI-CQA.

Contact

If you encounter any issues while using SCI-CQA, please feel free to contact us. We will update it promptly.

Downloads last month
2