FormulaNet
收藏资源简介:
在超过45k页的页面上,我们认为FormulaNet是具有内联公式标签的最大MFD数据集。我们的实验证明,在现有数据集上,内联和显示公式检测的标记质量大大提高。此外,我们为公式网提供了一个数学公式检测基线与0.754图。我们的数据集旨在帮助解决MFD任务,并可能实现新应用程序的开发,例如为视障屏幕阅读器用户在pdf中访问数学公式。
With over 45,000 pages, we consider FormulaNet to be the largest MFD dataset with inline formula annotations. Our experiments demonstrate that the annotation quality of inline and display formula detection has been significantly improved compared to existing datasets. Additionally, we have developed a mathematical formula detection baseline for FormulaNet, achieving a mean average precision (mAP) of 0.754 and providing the corresponding performance plot. Our dataset is designed to support the MFD task and enable the development of new applications, such as allowing visually impaired screen reader users to access mathematical formulas in PDF documents.




