SciCap: Generating Captions for Scientific Figures

TL;DR

This paper presents SciCap, a neural framework for automatically generating high-quality captions for scientific figures, supported by a large dataset and baseline models, addressing the challenge of improving figure communication in scientific papers.

Contribution

Introduction of SCICAP, a large-scale dataset of over two million scientific figures with captions, and baseline models for figure captioning in scientific literature.

Findings

Baseline models show potential for scientific figure captioning.

Generating accurate captions for diverse figure types remains challenging.

The dataset enables future research in scientific figure understanding.

Abstract

Researchers use figures to communicate rich, complex information in scientific papers. The captions of these figures are critical to conveying effective messages. However, low-quality figure captions commonly occur in scientific articles and may decrease understanding. In this paper, we propose an end-to-end neural framework to automatically generate informative, high-quality captions for scientific figures. To this end, we introduce SCICAP, a large-scale figure-caption dataset based on computer science arXiv papers published between 2010 and 2020. After pre-processing - including figure-type classification, sub-figure identification, text normalization, and caption text selection - SCICAP contained more than two million figures extracted from over 290,000 papers. We then established baseline models that caption graph plots, the dominant (19.2%) figure type. The experimental results showed both opportunities and steep challenges of generating captions for scientific figures.