PDFMathTranslate

Overview

PDFMathTranslate translates scientific PDF documents while preserving their layouts, including formulas, charts, tables of contents, and annotations. It provides command-line, browser-based GUI, Docker, and Zotero workflows, with support for multiple languages and translation services.

Key Features

• Preserves formulas, charts, table of contents, and annotations in translated PDFs.
• Supports multiple languages and diverse translation services, including Google, DeepL, Ollama, and OpenAI.
• Produces monolingual translated PDFs and bilingual dual PDFs.
• Offers command-line, interactive GUI, Docker, and Zotero plugin usage.
• Includes experimental OCR for image-only pages and an ultrafast mode for native-text PDFs.
• Provides Python and HTTP APIs for downstream development.

Use Cases

• Researchers and developers can translate scientific papers while retaining formulas, charts, and layout.
• Readers can generate bilingual dual PDFs alongside monolingual translated PDFs.
• Teams can run translation locally through the CLI or GUI, or deploy it with Docker.
• Developers can integrate PDF translation into downstream Python or HTTP API applications.

Getting Started

• Install with pip install pdf2zh, or use uv tool install --python 3.12 pdf2zh.
• Run pdf2zh document.pdf to translate a local PDF; output files are generated in the current working directory.
• Start the browser GUI with pdf2zh -i and open http://localhost:7860/ if it does not open automatically.
• Docker users can pull byaidu/pdf2zh and run it on port 7860.

Deployment & Requirements

• Python 3.11 to 3.12 is required for local Python installation.
• Docker deployment uses byaidu/pdf2zh and exposes port 7860.
• An alternative container image is available at ghcr.io/byaidu/pdfmathtranslate.
• The Windows app is distributed as a win64 zip; vc_redist.x64.exe may be needed if the app does not open.

Before You Adopt

• License: AGPL-3.0. Review its terms before using, modifying, or distributing the project.
• The AI model download can fail in some network environments.
• Precise mode is experimental and requires the pdf2zh_next submodule.