Quivr/OmniDocBench

OmniDocBench 是一个用于评估真实场景中多样化文档解析的基准数据集。该数据集包括各种文档类型、布局和语言，并对块级和跨度级文档元素进行丰富的注释。它还包括一个评估代码套件，用于端到端和单一模块的评估。数据集以 JSON 格式提供，并包括用于评估的相应图像和 PDF。README 还包括使用 Python 工具对数据集进行过滤和分层采样的说明。

OmniDocBench is a benchmark for diverse document parsing in real-world scenarios, with various document types, layouts, and languages, rich annotations for block-level and span-level document elements, and an evaluation code suite for end-to-end and single module evaluations. The dataset is available in JSON format and includes corresponding images and PDFs for evaluation. The README also includes instructions for using a Python tool for filtering and stratified sampling on the dataset.