reducto/ufocr

UFOCR是一个开放数据集，包含美国联邦调查局（FBI）和美国战争部解密的关于不明飞行物（UFO）、不明空中现象（UAP）以及外星调查的记录。这些文档通过Reducto技术完全解析为干净、结构化、适合大型语言模型（LLM）使用的文本。原始政府档案是OCR处理的难题，包括扫描的打字机页面、褪色的复写副本、手写边注、涂黑块、旋转扫描、多列备忘录和密集表格。数据集来源是FBI Vault和Department of War UFO Archive的公开解密文件，内容涵盖两个集合中的所有文档和图像，格式为结构化文本加布局元数据，可用于检索增强生成（RAG）、微调、搜索索引或分析。许可证为CC BY 4.0，语言为英语。

UFOCR is an open dataset of the FBIs and U.S. Department of Wars declassified records on UFOs, UAPs (unidentified aerial phenomena), and extraterrestrial investigations — fully parsed into clean, structured, LLM-ready text using Reducto. The original government archives are a notoriously difficult OCR challenge: scanned typewriter pages, faded carbon copies, handwritten margin notes, redacted blocks, rotated scans, multi-column memos, and dense tables. The dataset is sourced from public declassification releases from the FBI Vault and the Department of War UFO Archive, containing all documents and images from both collections, parsed end-to-end. The format is structured text plus layout metadata, ready for retrieval-augmented generation (RAG), fine-tuning, search indexing, or analysis. License: CC BY 4.0, Language: English.