aiptimizer/TurboOCR

TurboOCR, >200 img/s OmnidocBench. TensorRT FP16, PP-OCRv6, HTTP + gRPC

What it solves

TurboOCR is a high-performance GPU-accelerated document parser designed to convert images and PDFs into structured Markdown. It solves the speed bottleneck of traditional OCR and VLM (Vision Language Model) parsers, enabling the processing of hundreds of images per second on a single GPU while maintaining high accuracy for forms, receipts, and dense documents.

How it works

Built with C++, CUDA, and TensorRT, the project implements a multi-stream engine that runs a pipeline of specialized models rather than a single monolith. It utilizes PP-OCRv6 for text detection and recognition, PP-DocLayoutV3 for layout analysis, and specialized models like SLANet-Plus for tables (converting them to HTML) and PP-FormulaNet-S for mathematical formulas (converting them to LaTeX). The system is deployed as a server providing both HTTP and gRPC APIs, with TensorRT engines auto-built on the first startup to optimize performance for the specific GPU hardware.

Who it’s for

It is intended for developers and organizations needing industrial-scale document digitization, specifically those processing high volumes of receipts, forms, or academic papers where low latency and high throughput are critical.

Highlights

  • Extreme Throughput: Capable of processing up to 559 images per second on an RTX 5090 for receipts.
  • Structured Output: Goes beyond simple text extraction to provide layout analysis, reading order, HTML tables, and LaTeX formulas.
  • Flexible Accuracy Tiers: Offers tiny, small, and medium model tiers to let users trade speed for higher accuracy.
  • Native PDF Support: Renders and OCRs PDF pages in parallel with optional auto-rotation.
  • Broad Language Support: Covers Latin, Chinese, and Japanese by default, with additional support for Arabic, Cyrillic, Korean, Thai, and Greek.

Related

  • Project
    PaddlePaddle/PaddleOCR

    PaddleOCR is an open‑source OCR and document‑AI toolkit that converts images/PDFs into structured, LLM‑ready JSON or Markdown. It offers multilingual scene‑text models (PP‑OCRv6), a lightweight vision‑language parser (PaddleOCR‑VL‑1.6) for tables, formulas, and ancient scripts, and a structure‑aware pipeline (PP‑StructureV3). Models run on CPU, GPU, XPU, NPU or via ONNX/TensorRT, and integrate with RAG/agent platforms like Dify and RAGFlow.

  • Dispatch
    Hugging Face Accelerating Document AI

    Hugging Face provides a comprehensive guide to using open-source multimodal models to automate document classification, parsing, and visual question answering for enterprise workflows.

  • Project
    Yuliang-Liu/MonkeyOCRv2

    A visual-text foundation model for Document AI that provides a multilingual vision encoder and specialized models for document parsing and understanding.

  • Project
    arcships/light-ocr

    A fast, offline OCR library for Node.js and C++ that provides local text recognition for images and PDFs with built-in hardware acceleration.

  • Project
    Topkill/tianruoocr

    TianRuo OCR v6 is a Windows desktop app that captures screen text, runs OCR (online or offline via PaddleOCR/RapidOCR), and translates the result using many online services. It offers normal, silent, screenshot and clipboard‑listen modes, configurable shortcuts, and supports printed, table and handwritten text.