This repository contains code to export CRAFT (Character Region Awareness For Text Detection) model to ONNX format and run inference.

If you find this tool useful, consider becoming a sponsor for $1/month and get priority support!

• Docker 🐳
• High-resolution text images 🖼️

├── data/ # Add your high-res images here
├── onnx/ # ONNX model will be exported here
├── weights/ # craft_mlt_25k.pth model goes here
├── outputs/ # Detection results will be saved here
├── Dockerfile
├── onnx-export.py
└── onnx-inference.py

1. Add Images and download pth file 📸

   • Place your high-resolution text images in the data directory
   • Download pytorch model from here, and save it in weights folder

2. Build Docker Image 🔨

   docker build -t craft-onnx:latest .

3. Run Docker Container 🐋

   docker run -it craft-onnx:latest /bin/bash

4. Export ONNX Model 📤

   python3 onnx-export.py

   This will:

   • Take a sample image from data directory
   • Export ONNX graph to onnx folder

5. Run Inference 🔍

   python3 onnx-inference.py

   • Uses the exported ONNX model
   • Saves detection results in outputs directory

• Input size: 1280x960
• Optimized for high-resolution document images
• CPU-friendly inference

• Make sure images are readable and have sufficient resolution
• The model works best with clear, well-lit document images
• Check outputs directory for detection results

👉 My GitHub Sponsors link

• 🏢 Priority support for integrating this tool into your company's infrastructure
• 🛠️ Direct assistance with project-related issues and customizations
• 💡 Technical consultation for your specific use cases
• 🚀 Early access to new features and improvements
• ⭐ Recognition in our sponsors list

Your sponsorship helps me maintain the code, ensuring it remains a robust and reliable tool for the community. Every contribution, no matter how small, makes a difference!

MIT License

This repository's modifications and additions are licensed under MIT License.

• CRAFT Text Detector: MIT License
• PyTorch: BSD License
• ONNX Runtime: MIT License

This work builds upon:

• CRAFT-pytorch by CLOVA AI Research
• Other open source projects listed in requirements.txt