Installation & Usage

Overview¶

Source repository: https://github.com/RapidAI/RapidOCRAPI

rapidocr_api exposes RapidOCR through FastAPI and Uvicorn. It is intended as a simple HTTP wrapper. For production concurrency and process management, place it behind an appropriate deployment setup and consult the upstream repository.

Note

The API returns low-level OCR results. Applications should validate requests and transform the response for their own domain.

Version compatibility¶

`rapidocr_api`	OCR dependency
`v0.2.x`	`rapidocr>1.0.0,<3.0.0`
`v0.1.x`	Legacy `rapidocr_onnxruntime` package

Check the package release notes before combining newer RapidOCR versions with this API wrapper.

Installation¶

pip install rapidocr_api

Start the server¶

rapidocr_api -ip 0.0.0.0 -p 9005 -workers 2

The interactive FastAPI documentation is available at:

http://localhost:9005/docs

Model paths can be supplied through environment variables. For example on Linux:

export det_model_path=/path/to/detection.onnx
export rec_model_path=/path/to/recognition.onnx
rapidocr_api -ip 0.0.0.0 -p 9005 -workers 2

Call the API¶

cURL¶

curl -F image_file=@image.png http://localhost:9005/ocr

Python file upload¶

import requests

url = "http://localhost:9005/ocr"
image_path = "image.png"

with open(image_path, "rb") as image_file:
    files = {"image_file": (image_path, image_file, "image/png")}
    response = requests.post(url, files=files, timeout=60)

response.raise_for_status()
print(response.json())

Base64 input¶

import base64
from pathlib import Path

import requests

encoded = base64.b64encode(Path("image.png").read_bytes())
response = requests.post(
    "http://localhost:9005/ocr",
    data={"image_data": encoded},
    timeout=60,
)

response.raise_for_status()
print(response.json())

Response format¶

When text is detected, the response maps result indexes to recognized text, box coordinates and confidence scores:

{
  "0": {
    "rec_txt": "RapidOCR",
    "dt_boxes": [[10, 20], [180, 20], [180, 60], [10, 60]],
    "score": "0.98"
  }
}

If no text is detected, the API returns an empty object: {}.

For Docker examples, historical package layouts and additional request flags, see the complete Chinese API reference.