jonathan-roberts1/SciFIBench

---
task_categories:
- question-answering
tags:
- science
pretty_name: Scientific Figure Interpretation Benchmark
size_categories:
- 1k<n<10k
language:
- en
---

# Dataset Card for SciFIBench

## Dataset Description

- **Homepage:** [https://github.com/jonathan-roberts1/SciFIBench](https://github.com/jonathan-roberts1/SciFIBench)
- **Paper:** [SciFIBench: Benchmarking Large Multimodal Models for Scientific Figure Interpretation](https://arxiv.org/pdf/2405.08807)

### Dataset Summary
The SciFIBench (Scientific Figure Interpretation Benchmark) contains 1000 multiple-choice scientific figure interpretation questions covering two tasks. Task 1:
Figure -> Caption involves selecting the most appropriate caption given a figure; Task 2: Caption -> Figure involves the opposite -- selecting the most appropriate
figure given a caption. This benchmark was curated from the SciCap dataset, using adversarial filtering to obtain hard negatives. Human verification has been performed
on each question to ensure high-quality,
answerable questions.

### Source Data

More information regarding the source data can be found at: https://github.com/tingyaohsu/SciCap

### Dataset Curators

This dataset was curated by Jonathan Roberts, Kai Han, Neil Houlsby, and Samuel Albanie

### Citation Information
```
@article{roberts2024scifibench,
title={SciFIBench: Benchmarking Large Multimodal Models for Scientific Figure Interpretation},
author={Jonathan Roberts and Kai Han and Neil Houlsby and Samuel Albanie},
year={2024},
journal={arXiv preprint arXiv:2405.08807},
}
```

原始信息汇总

数据集概述

基本信息

• 名称: Scientific Figure Interpretation Benchmark (SciFIBench)
• 任务类别: 问答 (question-answering)
• 标签: 科学 (science)
• 大小范围: 1k<n<10k
• 语言: 英语 (en)

数据集描述

• 概述: SciFIBench包含1000个多选科学图表解释问题，覆盖两个任务。任务1：图表->标题，涉及选择给定图表的最合适标题；任务2：标题->图表，涉及选择给定标题的最合适图表。
• 来源: 该基准从SciCap数据集精选，使用对抗性过滤获取困难负例，并通过人工验证确保问题的高质量和可回答性。

数据集管理

• 管理员: Jonathan Roberts, Kai Han, Neil Houlsby, Samuel Albanie

引用信息

@article{roberts2024scifibench, title={SciFIBench: Benchmarking Large Multimodal Models for Scientific Figure Interpretation}, author={Jonathan Roberts and Kai Han and Neil Houlsby and Samuel Albanie}, year={2024}, journal={arXiv preprint arXiv:2405.08807}, }