opendatalab/mineru

MinerU is an open-source document parsing tool that converts complex documents, particularly PDFs, but also Office file formats, into clean Markdown or JSON output that AI systems can easily process. The problem it addresses is that PDFs and other document formats are notoriously difficult to work with programmatically. They may contain multi-column layouts, embedded tables, mathematical formulas, images with text, headers and footers, and footnotes, all of which a naive text extractor would mix up or miss entirely. MinerU produces structured output that preserves the logical reading order and document structure. Under the hood MinerU applies layout analysis, using computer vision models to detect regions on each page and classify them as paragraphs, tables, figures, headings, and so on, before extracting and ordering the content. It also integrates OCR (optical character recognition, the process of reading text from images) for handling scanned documents or embedded images containing text. The extracted content is output as Markdown with proper headings, table formatting, and code blocks, making it immediately usable for feeding into large language models for question-answering, summarization, or indexing in a retrieval system. It is available as a Python library installable via pip, as a web application on HuggingFace and ModelScope, and can be run in Google Colab notebooks. You would use MinerU when building an AI pipeline that needs to ingest existing PDF documents, research papers, reports, or Office files. The tech stack is Python with deep learning models for layout analysis, available via pip.