lance-format/docvqa-lance
收藏资源简介:
DocVQA (Lance Format)是DocVQA数据集的Lance格式版本,主要用于文档图像上的视觉问答(VQA)。文档图像包括行业/政府扫描件、多页报告、表单、收据等。数据集包含验证集和测试集,分别有5,349和5,188行数据。数据集的schema详细列出了各列的名称、类型和说明,包括图像、问题、答案、文档ID等信息。此外,还提供了预构建的索引、快速开始的代码示例以及按问题类型过滤的示例。数据集的来源和许可证信息也明确给出,DocVQA采用MIT许可证,底层文档来自UCSF Industry Documents Library。
DocVQA (Lance Format) is a Lance-formatted version of the DocVQA dataset, designed for Visual Question Answering (VQA) over document images. The document images include industry/government scans, multi-page reports, forms, receipts, etc. The dataset contains validation and test splits with 5,349 and 5,188 rows respectively. The schema of the dataset details the column names, types, and notes, including image, question, answer, document ID, etc. Additionally, pre-built indices, quick start code examples, and filtering by question type are provided. The source and license information is also clearly stated, with DocVQA released under the MIT license and the underlying documents sourced from the UCSF Industry Documents Library.




