PaddleOCR-VL Service Deployment & API Usage Example:

PaddleOCR open-source project GitHub address, this service is built based on the PaddleOCR-VL model from this open-source project.

Version Information: The current version on the PaddleOCR official website corresponds to PaddleX version 3.3.12 and PaddlePaddle version 3.2.1.

1. Introduction to PaddleOCR-VL

PaddleOCR-VL is an advanced and efficient document analysis model designed for element recognition within documents. Its core component, PaddleOCR-VL-0.9B, is a compact yet powerful Vision-Language Model (VLM) composed of a NaViT-style dynamic resolution visual encoder and an ERNIE-4.5-0.3B language model, enabling precise element recognition. The model supports 109 languages and excels at identifying complex elements such as text, tables, formulas, and charts, all while maintaining extremely low resource consumption. Comprehensive evaluations on widely-used public benchmarks and internal datasets demonstrate that PaddleOCR-VL achieves state-of-the-art (SOTA) performance in both page-level document analysis and element-level recognition. It significantly outperforms existing pipeline-based solutions, other multimodal document analysis approaches, and advanced general-purpose multimodal large models, while offering faster inference speed. These advantages make it highly suitable for real-world deployment.

Key Metrics:

Core Features:

1. Compact and Powerful Vision-Language Model Architecture:
   We propose a novel vision-language model specifically designed for resource-efficient inference and outstanding element recognition. By integrating a NaViT-style dynamic high-resolution visual encoder with a lightweight ERNIE-4.5-0.3B language model, we significantly enhance recognition capability and decoding efficiency. This integration maintains high accuracy while reducing computational requirements, making it ideal for efficient and practical document processing applications.
2. SOTA Performance in Document Analysis:
   PaddleOCR-VL achieves state-of-the-art performance in both page-level document parsing and element-level recognition. It significantly outperforms traditional pipeline-based solutions and demonstrates competitive strength against leading vision-language models (VLMs) in document analysis. Furthermore, it excels at identifying complex document elements such as text, tables, formulas, and charts, making it suitable for challenging content types including handwritten text and historical documents. This versatility enables it to handle a wide range of document types and scenarios.
3. Multilingual Support:
   PaddleOCR-VL supports 109 languages, covering the major global languages such as Chinese, English, Japanese, Latin, and Korean, as well as languages with diverse scripts and structures like Russian (Cyrillic), Arabic, Hindi (Devanagari), and Thai. This extensive language coverage greatly enhances the system's applicability in multilingual and global document processing scenarios.

The following diagram illustrates the overall workflow of PaddleOCR-VL:

2. API Quota Rules and Error Code Description

Please refer to the documentation.

3. Service Call Example (python)

# Please make sure the requests library is installed
# pip install requests
import base64
import os
import requests

# Please visit https://aistudio.baidu.com/paddleocr/task to obtain the API_URL and TOKEN in the API call example.
API_URL = "<your url>"
TOKEN = "<access token>"

file_path = "<local file path>"

with open(file_path, "rb") as file:
    file_bytes = file.read()
    file_data = base64.b64encode(file_bytes).decode("ascii")

headers = {
    "Authorization": f"token {TOKEN}",
    "Content-Type": "application/json"
}

required_payload = {
    "file": file_data,
    "fileType": <file type>,  # For PDF documents, set `fileType` to 0; for images, set `fileType` to 1
}

optional_payload = {
    "useDocOrientationClassify": False,
    "useDocUnwarping": False,
    "useChartRecognition": False,
}

payload = {**required_payload, **optional_payload}


response = requests.post(API_URL, json=payload, headers=headers)
print(response.status_code)
assert response.status_code == 200
result = response.json()["result"]

output_dir = "output"
os.makedirs(output_dir, exist_ok=True)

for i, res in enumerate(result["layoutParsingResults"]):
    md_filename = os.path.join(output_dir, f"doc_{i}.md")
    with open(md_filename, "w", encoding="utf-8") as md_file:
        md_file.write(res["markdown"]["text"])
    print(f"Markdown document saved at {md_filename}")
    for img_path, img in res["markdown"]["images"].items():
        full_img_path = os.path.join(output_dir, img_path)
        os.makedirs(os.path.dirname(full_img_path), exist_ok=True)
        img_bytes = requests.get(img).content
        with open(full_img_path, "wb") as img_file:
            img_file.write(img_bytes)
        print(f"Image saved to: {full_img_path}")
    for img_name, img in res["outputImages"].items():
        img_response = requests.get(img)
        if img_response.status_code == 200:
            # Save image to local
            filename = os.path.join(output_dir, f"{img_name}_{i}.jpg")
            with open(filename, "wb") as f:
                f.write(img_response.content)
            print(f"Image saved to: {filename}")
        else:
            print(f"Failed to download image, status code: {img_response.status_code}")

Main operations provided by the service:

• The HTTP request method is POST.
• Both the request body and response body are in JSON format (JSON object).
• When the request is processed successfully, the response status code is 200, and the response body contains the following properties:

• When the request is not processed successfully, the response body contains the following properties:

Main operations provided by the service are as follows:

• infer

Performs document analysis.

POST /layout-parsing

4. Request Parameter Description

Serving:
  extra:
    max_num_input_imgs: null

Serving:
  visualize: False

• When the request is processed successfully, the result field in the response body has the following properties:

Each element in layoutParsingResults is an object with the following properties:

markdown is an object with the following properties:

For details on the returned data structure and field descriptions, please refer to the documentation.

Note: If you encounter any issues during use, please feel free to submit feedback in the issue section.