PaddleOCR ONNX 模型

PaddleOCR PP-OCRv5 模型已转换为 ONNX 格式，用于高效的 OCR 推理。

English | 中文说明

中文说明

这是 PaddleOCR PP-OCRv5 模型的 ONNX 格式版本，用于高效率的文字识别推理。

模型文件说明

快速使用

from huggingface_hub import hf_hub_download

# 下载文字检测模型
det_model_path = hf_hub_download(
    repo_id="marsena/paddleocr-onnx-models",
    filename="PP-OCRv5_server_det_infer.onnx"
)

# 下载文字识别模型  
rec_model_path = hf_hub_download(
    repo_id="marsena/paddleocr-onnx-models",
    filename="PP-OCRv5_server_rec_infer.onnx"
)

一次下载所有模型

from huggingface_hub import snapshot_download

# 下载所有模型文件到本地文件夹
snapshot_download(
    repo_id="marsena/paddleocr-onnx-models",
    local_dir="./paddleocr_onnx"
)

支持的语言和文字类型

• 语言支持: 简体中文、繁体中文、英文、日文
• 文字类型: 印刷体、手写体、竖排文字、旋转文字、弯曲文字
• 输入格式: 图片文件 (JPEG, PNG)
• 运行环境: 支持 CPU 和 GPU 推理

English

Model Files

Source Models

These ONNX models are converted from official PaddlePaddle PP-OCRv5 models:

• Detection Model: PaddlePaddle/PP-OCRv5_server_det
• Recognition Model: PaddlePaddle/PP-OCRv5_server_rec
• Official Documentation: PP-OCRv5 Introduction

Usage

Download Specific Model

from huggingface_hub import hf_hub_download

# Download detection model
det_model_path = hf_hub_download(
    repo_id="marsena/paddleocr-onnx-models",
    filename="PP-OCRv5_server_det_infer.onnx"
)

# Download recognition model  
rec_model_path = hf_hub_download(
    repo_id="marsena/paddleocr-onnx-models",
    filename="PP-OCRv5_server_rec_infer.onnx"
)

Download All Models

from huggingface_hub import snapshot_download

# Download all model files to local directory
snapshot_download(
    repo_id="marsena/paddleocr-onnx-models",
    local_dir="./paddleocr_onnx"
)

ONNX Runtime Inference

import onnxruntime as ort
import numpy as np

# Load model
session = ort.InferenceSession("PP-OCRv5_server_det_infer.onnx")

# Run inference
input_name = session.get_inputs()[0].name
output = session.run(None, {input_name: input_data})

Model Specifications

• Languages: Simplified Chinese, Traditional Chinese, English, Japanese
• Text Types: Printed text, handwriting, vertical text, rotated text
• Input Format: Images (JPEG, PNG)
• Output Format: Bounding boxes + recognized text
• Runtime: ONNX Runtime 1.16+
• Hardware: CPU and GPU inference supported

License

These models follow the Apache License 2.0, consistent with the original PaddleOCR project.

• PaddleOCR Repository: https://github.com/PaddlePaddle/PaddleOCR
• License Details: Apache License 2.0

Conversion Information

• Conversion Tool: Paddle2ONNX
• ONNX Version: 1.12+
• Source Framework: PaddlePaddle 2.5+
• Conversion Date: January 2025

Citation

If you use these models in your research, please cite the original PaddleOCR paper:

@misc{paddleocr2020,
    title={PaddleOCR: Awesome multilingual OCR toolkits},
    author={PaddlePaddle Authors},
    howpublished = {\url{https://github.com/PaddlePaddle/PaddleOCR}},
    year={2020}
}

Issues

For model usage issues, please report to the original PaddleOCR repository:

• PaddleOCR Issues: https://github.com/PaddlePaddle/PaddleOCR/issues