A Torch-based deep learning project for breaking CAPTCHA systems using CNN and RNN architectures.
Captcha is a deep learning project that uses Torch to break CAPTCHA systems through neural network models. It implements convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to recognize and solve CAPTCHA challenges, demonstrating vulnerabilities in these security systems while providing educational code examples.
Machine learning researchers, security analysts, and developers interested in deep learning applications for computer vision and sequence recognition tasks.
It offers practical, runnable implementations of CNN and RNN models specifically for CAPTCHA breaking, with documented accuracy results and clear examples for educational and research purposes.
Breaking captchas using torch
Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.
Provides runnable Lua scripts and Jupyter notebooks that allow direct experimentation with CNN and RNN models on CAPTCHA data, making it accessible for learning.
Shows a CNN-RNN architecture achieving 96% accuracy on SimpleCaptcha, offering practical insights into combining vision and sequence models for improved performance.
Documents specific accuracy scores (92% for CNN, 96% for CNN-RNN) that help benchmark model effectiveness and understand trade-offs in deep learning approaches.
Demonstrates how deep learning can break CAPTCHAs, providing valuable insights into vulnerabilities for security analysts and researchers.
Built on Torch, which is largely superseded by PyTorch, leading to potential compatibility issues and a steeper learning curve for modern developers.
Only tested on SimpleCaptcha datasets with RNN models at 50% accuracy, so it may not generalize to more robust or diverse CAPTCHA systems used today.
The README is minimal, lacking detailed setup instructions, model architecture explanations, or guidance for extending the project beyond basic examples.