PP-OCRv5 ONNX Export Pipeline

Streamlined pipeline for exporting PP-OCRv5 models to ONNX format. Contains a GH action workflow and automated releases.

🚀 Quick Start

# Prerequisites: Install just command runner
# curl --proto '=https' --tlsv1.2 -sSf https://just.systems/install.sh | bash

# Complete pipeline - one command does it all
just quick-start

# Or step by step
just deps        # Install Python dependencies
just export-all  # Export all 4 models to onnx_models/

Features

• ✅ Official Method: Uses PaddleOCR's official tools/export_model.py
• ✅ Auto Setup: Downloads PaddleOCR repo and pretrained models automatically
• ✅ Fixed Issues: Resolves path and JSON format problems from original scripts
• ✅ CI/CD Ready: GitHub Action included for automated releases
• ✅ Proper Naming: Preserves original PP-OCRv5 model names

Available Models

Model	Type	Size	Description
PP-OCRv5_server_det	Detection	~88MB	High accuracy detection model
PP-OCRv5_server_rec	Recognition	~84MB	High accuracy recognition model
PP-OCRv5_mobile_det	Detection	~5MB	Optimized mobile detection
PP-OCRv5_mobile_rec	Recognition	~16MB	Optimized mobile recognition

Commands

just info           # Show pipeline information
just deps           # Install Python dependencies
just setup          # Setup PaddleOCR repository and download models
just export-all     # Export all models
just validate       # Validate exported ONNX models
just clean          # Clean intermediate files
just clean-all      # Full cleanup
just quick-start    # Complete setup and export

Individual Model Export

just export-server-det    # PP-OCRv5_server_det.onnx
just export-server-rec    # PP-OCRv5_server_rec.onnx
just export-mobile-det    # PP-OCRv5_mobile_det.onnx
just export-mobile-rec    # PP-OCRv5_mobile_rec.onnx

GitHub Actions

The repository includes a GitHub Action that:

• Automatically exports models on tag pushes (v*)
• Can be triggered manually with workflow_dispatch
• Uploads models as release artifacts
• Creates GitHub releases with model files

Manual Trigger

1. Go to Actions tab in GitHub
2. Select "Export PP-OCRv5 Models"
3. Click "Run workflow"
4. Optionally specify a release tag

Automatic Release

git tag v1.0.0
git push origin v1.0.0

Using Exported Models

import onnxruntime as ort
import numpy as np

# Load detection model
det_session = ort.InferenceSession('onnx_models/PP-OCRv5_server_det.onnx')

# Load recognition model
rec_session = ort.InferenceSession('onnx_models/PP-OCRv5_server_rec.onnx')

# Run inference
det_result = det_session.run(None, {"x": input_image})
rec_result = rec_session.run(None, {"x": cropped_text})

File Structure

├── scripts/
│   └── export.py             # Fixed export script
├── .github/workflows/
│   └── export-models.yml     # GitHub Action
├── justfile                  # Command definitions
├── requirements.txt          # Python dependencies
└── README.md                 # This file

Requirements

• Python 3.8+
• just command runner
• Git

Dependencies are automatically installed with just deps.

Technical Details

• Export Method: Official PaddleOCR tools/export_model.py
• Format Support: Handles PaddlePaddle 3.x JSON export format
• Path Handling: Uses absolute paths to avoid construction issues
• ONNX Version: Auto-selected opset (typically 14)
• Optimization: Includes onnxoptimizer for model optimization

License

This project follows the same license as the original PaddleOCR project.

Credits

• PaddleOCR - Original OCR framework
• PaddlePaddle - Deep learning framework