datajuicer/VeriSciQA

---
license: cc-by-sa-4.0
task_categories:
- visual-question-answering
language:
- en
tags:
- scientific-vqa
- vision-language
- scientific-figures
- multi-choice-qa
pretty_name: VeriSciQA
size_categories:
- 10K<n<100K
dataset_info:
features:
- name: image
dtype: image
- name: question
dtype: string
- name: options
sequence: string
- name: answer
dtype: string
- name: caption
dtype: string
- name: figure_type
dtype: string
- name: image_label
dtype: string
- name: section
dtype: string
- name: domain
dtype: string
- name: question_type
dtype: string
---

# VeriSciQA: An Auto-Verified Dataset for Scientific Visual Question Answering

**Paper**: [VeriSciQA: An Auto-Verified Dataset for Scientific Visual Question Answering](https://arxiv.org/abs/2511.19899)

## Dataset Description

VeriSciQA is a large-scale, high-quality dataset for Scientific Visual Question Answering (SVQA), containing 20,272 QA pairs spanning 20 scientific domains, 12 figure types, and 5 question types. The dataset is constructed using a Cross-Modal Verification framework that generates QA pairs from figure-citing paragraphs and verifies their visual grounding against the corresponding figures, leveraging cross-modal consistency to filter out erroneous pairs.

### Key Features

- **20,272 QA pairs** covering diverse scientific figures from peer-reviewed papers
- **20 scientific domains**: including Computer Science, Physics, Mathematics, Biology, etc.
- **12 figure types**: Line plots, Bar charts, Scatter plots, Diagrams, Heatmaps, and more
- **5 question types**: Comparative, Compositional, Descriptive, Relational, and Structural
- **Multiple-choice format**: 4 options per question
- **Auto-verified**: Cross-modal consistency checks to minimize errors

### Dataset Statistics

| Metric | Value |
|--------|-------|
| Total QA pairs | 20,272 |
| Scientific domains | 20 |
| Figure types | 12 |
| Question types | 5 |

## Usage

```python
from datasets import load_dataset

dataset = load_dataset("datajuicer/VeriSciQA", split="train")
print(dataset[0])
```

## Dataset Structure

### Data Fields

Each example in the dataset contains:

- `image`: (PIL Image) The scientific figure
- `question`: (string) The question about the figure
- `options`: (list of 4 strings) Multiple-choice options
- `answer`: (string) The correct answer choice (A/B/C/D)
- `caption`: (string) Original figure caption from the paper
- `figure_type`: (string) Type of figure (e.g., "Line Plot", "Bar Chart", "Diagram")
- `image_label`: (string) Figure label from the original paper
- `section`: (string) Relevant section text from the paper providing context
- `domain`: (string) Scientific domain (e.g., "cs", "physics", "math", "cond-mat")
- `question_type`: (string) Question type (Comparative, Compositional, Descriptive, Relational, Structural)

### Data Example

```json
{
"image": "cond-mat0506675_2.jpg",
"question": "In the figure, what is the direction of the angular velocity of the flagellar bundle relative to the cell body?",
"options": [
"Clockwise around the y-axis",
"Counter-clockwise around the y-axis",
"Counter-clockwise around the z-axis",
"Clockwise around the x-axis"
],
"answer": "B",
"caption": "Set-up and notations for the mechanical model of {E. coli} swimming near a solid surface.",
"figure_type": "Diagram",
"image_label": "setup",
"section": "We model the bacterium as a single, left-handed rigid helix attached to a spherical body...",
"domain": "cond-mat",
"question_type": "Relational"
}
```

## Citation

If you use this dataset in your research, please cite:

```bibtex
@article{verisciqa2025,
title={VeriSciQA: An Auto-Verified Dataset for Scientific Visual Question Answering},
author={Li, Yuyi and Chen, Daoyuan and Wang, Zhen and Lu, Yutong and Li, Yaliang},
journal={arXiv preprint arXiv:2511.19899},
year={2025},
url={https://arxiv.org/abs/2511.19899}
}
```

## License

This dataset is released under the **CC BY-SA 4.0** license.