GOT-OCR 2.0 Microservice 🚀

State-of-the-art OCR microservice powered by Transformers and FastAPI, supporting multiple OCR tasks and formats.

📌 Key Features

• Multi-Task OCR Support
  • Plain Text Extraction
  • Formatted Text (LaTeX/Markdown)
  • Region-Specific OCR (Box/Color)
  • Multi-Page Document Processing
  • Sheet Music & Math Formula Recognition
• Input Formats: JPEG, PNG, TIFF
• Output Formats: Raw Text, HTML, SVG
• GPU Acceleration Support (CUDA)

🛠 Installation

Prerequisites

• Python 3.9+
• CUDA 11.7+ (For GPU acceleration)
• Docker 20.10+ (Optional)

Without Docker

# Clone repository
git clone https://github.com/iammuhammadnoumankhan/FastAPI-GOT-OCR-2-Transformers.git
cd got-ocr-service

# Install dependencies
pip install -r requirements.txt

# Start service (CPU)
uvicorn main:app --host 0.0.0.0 --port 8000

# Start with GPU
CUDA_VISIBLE_DEVICES=0 uvicorn main:app --host 0.0.0.0 --port 8000

With Docker

# Build the Docker image:
docker build -t got-ocr-service .

# Run the container:
docker run -p 8000:8000 --gpus all got-ocr-service

🚀 Usage

Supported Tasks

Task	Parameters	Description
Plain Text OCR	None	Basic text extraction
Format Text OCR	ocr_type=format	Structured text output
Fine-grained OCR (Box)	ocr_box=[x1,y1,x2,y2]	Region-specific extraction
Fine-grained OCR (Color)	ocr_color=red/green/blue	Color-based extraction
Multi-crop OCR	None	Multiple region processing
Multi-page OCR	Multiple images	Document processing

API Endpoints

Endpoint	Method	Description
/process	POST	Main OCR processing
/results/{result_id}	GET	Retrieve HTML results
/docs	GET	Interactive API documentation

📋 Example Requests

1. Basic Text Extraction

curl -X POST "http://localhost:8000/process" \
  -F "task=Plain Text OCR" \
  -F "images=@document.jpg"

2. Math Formula Recognition

curl -X POST "http://localhost:8000/process" \
  -F "task=Format Text OCR" \
  -F "ocr_type=format" \
  -F "images=@equation.png"

3. Multi-Page Processing

curl -X POST "http://localhost:8000/process" \
  -F "task=Multi-page OCR" \
  -F "images=@page1.pdf" \
  -F "images=@page2.pdf"

4. Color-Based Extraction

curl -X POST "http://localhost:8000/process" \
  -F "task=Fine-grained OCR (Color)" \
  -F "ocr_color=red" \
  -F "images=@highlighted_text.png"

📚 API Documentation

Interactive Swagger documentation available at: http://localhost:8000/docs

Build & Run

# Build the Docker image:
docker build -t got-ocr-service .

# Run the container:
docker run -p 8000:8000 --gpus all got-ocr-service

📜 License

APACHE 2.0 License

---

Note: Replace localhost:8000 with your domain in production deployments. For large-scale usage, consider adding:

• Redis caching
• Load balancing
• Rate limiting

More CURLs Commands:

Here are the curl commands for all the supported use cases and types in the GOT-OCR 2.0 FastAPI microservice:

---

1. Plain Text OCR

Extracts plain text from an image.

curl -X POST "http://localhost:8000/process" \
  -F "task=Plain Text OCR" \
  -F "images=@document.jpg"

---

2. Formatted Text OCR

Extracts formatted text (e.g., LaTeX, Markdown) from an image.

curl -X POST "http://localhost:8000/process" \
  -F "task=Format Text OCR" \
  -F "ocr_type=format" \
  -F "images=@formatted_document.png"

---

3. Fine-grained OCR (Box)

Extracts text from a specific bounding box region in the image.

curl -X POST "http://localhost:8000/process" \
  -F "task=Fine-grained OCR (Box)" \
  -F "ocr_box=[100,100,300,300]" \
  -F "images=@image_with_regions.jpg"

---

4. Fine-grained OCR (Color)

Extracts text from regions highlighted with a specific color.

curl -X POST "http://localhost:8000/process" \
  -F "task=Fine-grained OCR (Color)" \
  -F "ocr_color=red" \
  -F "images=@color_highlighted_image.png"

---

5. Multi-crop OCR

Processes multiple cropped regions of an image.

curl -X POST "http://localhost:8000/process" \
  -F "task=Multi-crop OCR" \
  -F "images=@multi_crop_image.jpg"

---

6. Multi-page OCR

Processes multiple pages of a document.

curl -X POST "http://localhost:8000/process" \
  -F "task=Multi-page OCR" \
  -F "images=@page1.png" \
  -F "images=@page2.png" \
  -F "images=@page3.png"

---

7. Sheet Music OCR

Processes sheet music and generates formatted output.

curl -X POST "http://localhost:8000/process" \
  -F "task=Format Text OCR" \
  -F "ocr_type=format" \
  -F "images=@sheet_music.png"

---

8. Math Formula OCR

Extracts mathematical formulas from an image.

curl -X POST "http://localhost:8000/process" \
  -F "task=Format Text OCR" \
  -F "ocr_type=format" \
  -F "images=@math_formula.png"

---

9. Table and Chart OCR

Extracts structured data from tables and charts.

curl -X POST "http://localhost:8000/process" \
  -F "task=Format Text OCR" \
  -F "ocr_type=format" \
  -F "images=@table_chart.png"

---

10. Batch Processing

Process multiple images in a single request.

curl -X POST "http://localhost:8000/process" \
  -F "task=Plain Text OCR" \
  -F "images=@image1.jpg" \
  -F "images=@image2.png" \
  -F "images=@image3.tiff"

---

11. Retrieve HTML Results

After processing, retrieve the HTML-rendered result using the result_id.

curl -X GET "http://localhost:8000/results/{result_id}"

Replace {result_id} with the ID returned in the response from the /process endpoint.

---

12. Error Handling Examples

Missing Image

curl -X POST "http://localhost:8000/process" \
  -F "task=Plain Text OCR"

Response:

{
  "detail": "No image provided"
}

Invalid Task

curl -X POST "http://localhost:8000/process" \
  -F "task=Invalid Task" \
  -F "images=@document.jpg"

Response:

{
  "detail": "Invalid task specified"
}

Invalid Color

curl -X POST "http://localhost:8000/process" \
  -F "task=Fine-grained OCR (Color)" \
  -F "ocr_color=purple" \
  -F "images=@image.jpg"

Response:

{
  "detail": "Invalid color specified"
}

---

Notes:

1. Replace http://localhost:8000 with your actual server URL if deployed elsewhere.
2. Ensure the image files (@document.jpg, @formatted_document.png, etc.) exist in the directory where you run the curl command.
3. For multi-file uploads, use multiple -F "images=@file" fields.
4. The ocr_box parameter should be in the format [x1,y1,x2,y2].
5. The ocr_color parameter supports only red, green, or blue.

These commands cover all the supported use cases and types for the GOT-OCR 2.0 microservice.