This project addresses text selection CAPTCHA automation, offering a solution for researchers and developers. It provides a fast, accurate, and low-resource method for recognizing and ordering characters, enabling automated interaction with CAPTCHA-protected systems.

How It Works

The system uses YOLOv5 for high-speed, accurate character detection and localization. A Siamese network then performs image matching against a character library to determine the correct selection order, overcoming traditional sorting difficulties. This combination ensures efficient and reliable CAPTCHA resolution.

Quick Start & Requirements

Installation requires pip install -r requirements.txt. Basic usage is via python demo.py, while a service can be launched with python service.py (API docs at http://127.0.0.1:8000/docs#). Pre-trained models are restricted to Windows with Python 3.6, 3.8, or 3.10. Custom ONNX models bypass these platform limitations. Links to YOLOv5 and Siamese-pytorch repositories are provided.

Highlighted Details

• Achieves recognition speeds from 163ms to 500ms with up to 96% accuracy on specific CAPTCHA types.
• Demonstrates effective training on small datasets (e.g., 300 images).
• Designed for low resource consumption, runnable on low-configuration servers.
• Includes functionality for "消消乐" CAPTCHAs and supports custom ONNX model integration.

Maintenance & Community

Contact MgArcher via email (yj970814@163.com) for custom needs or inquiries. Associated with "道满PythonAI," a Chinese AI and Python educational platform. Users are encouraged to raise issues.

Licensing & Compatibility

The project is for learning and exchange only, prohibiting illegal or commercial use. A specific open-source license is not detailed, but usage restrictions imply a non-commercial, research context. Pre-trained models are Windows-limited; custom ONNX models offer broader compatibility.

Limitations & Caveats

Platform dependency is a key limitation: pre-trained models and direct execution are restricted to Windows with specific Python versions. Models are trained for particular CAPTCHA types, requiring custom training for others. Commercial use is expressly forbidden.