# jingsongliujing/OnnxOCR

基于PaddleOCR重构，并且脱离PaddlePaddle深度学习训练框架的轻量级OCR，推理速度超快   ——  A lightweight OCR system based on PaddleOCR, decoupled from the PaddlePaddle deep learning training framework, with ultra-fast inference speed.

Repository: https://github.com/jingsongliujing/OnnxOCR
Canonical: https://ross.abutalabs.com/products/onnxocr
Language: Python
License: Apache-2.0
License Family: permissive
Topics: document-layout-analysis, license-plate-recognition, ocr, onnx-models, onnxocr, onnxruntime, general-ocr, structured-ocr, tablerecognition
Last push: 2026-06-11T14:36:11+00:00

## Health v2 (maintenance only)
Score: 74/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 87, release rhythm 54, longevity 82
- inputs: {"age_days": 1152, "days_push": 83, "days_rel": 99, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1860, forks 200 (observed 2026-08-28T04:05:45.327300+00:00)

## What it is
A lightweight multilingual OCR library rebuilt from PaddleOCR models to run on ONNXRuntime, removing the PaddlePaddle dependency for fast inference. It also bundles ONNX-based license plate recognition, table recognition, document layout analysis with Markdown export, and an optional Qwen3.5-2B ONNX information-extraction workflow.

## Use cases
- extract text from images without installing paddlepaddle
- run OCR on ARM or x86 devices with onnxruntime
- recognize license plates from photos
- convert scanned documents to markdown with layout analysis
- recognize tables in images
- extract structured fields from ID cards using OCR plus a small LLM
- deploy fast multilingual OCR including Chinese, English, and Japanese

## When to choose
- you need PaddleOCR-quality OCR but cannot ship the PaddlePaddle framework
- you need cross-architecture deployment (ARM/x86) via ONNX models
- you want OCR plus table, layout, or license-plate recognition in one Python package
- inference speed and lightweight deployment matter more than training capability

## When to avoid
- you need to train or fine-tune OCR models yourself
- you need OCR languages outside the supported multilingual model set
- you require a managed cloud OCR API rather than a self-run Python library

## Facets
- artifact type: library
- maturity: active
- function: ocr, image-processing, machine-learning, llm-inference, http-server
- domain: computer-vision, image-processing, developer-tools
- platform: python, cross-platform, cli
- tags: onnxruntime, paddleocr, pp-ocrv5, table-recognition, layout-analysis, license-plate-recognition, text-detection, text-recognition, onnx-models, document-parsing, natural-language-processing

## Member repositories
- jingsongliujing/OnnxOCR (main) score 74

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:45.327300+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:15:55.228650+00:00, confidence not recorded.
  - readme: https://github.com/jingsongliujing/OnnxOCR (fetched 2026-08-28T04:05:45.327300+00:00, sha 8c4e3f795ab4)
  - registry_pypi: https://pypi.org/pypi/onnxocr/json (fetched 2026-08-29T10:55:02.245735+00:00, sha f82722a543f0)
- Data as of 2026-08-30T08:39:29.467469+00:00.