How can I adapt this model for other websites' captchas?

You need to retrain the neural network on a new dataset specific to that website's captcha format. The code is adjustable, but it requires machine learning expertise and significant data collection, as the model is optimized for IRCTC's 50x170 image size.

Is this better than using Tesseract OCR for captchas?

Yes, for IRCTC captchas, this deep learning approach achieves 98% accuracy, outperforming traditional OCR tools like Tesseract that struggle with distorted text. However, it demands more computational resources and setup time.

What hardware is required to run this project effectively?

You need a GPU, preferably an NVIDIA one like Titan 780, for efficient training. Without a GPU, training times become prohibitively long, as highlighted in the README's requirements section.

How to train the model on my own captcha dataset?

Install Torch and dependencies, prepare your dataset in the format shown in the data folder, and run scripts like mainr.lua or mainaug.lua while adjusting parameters like batch size and data folder paths in main.lua.

What are the ethical issues with using this code?

While it showcases security vulnerabilities for research, it can be misused for automated ticket booking, which violates IRCTC's terms. Use it responsibly for educational purposes and consider the legal implications.

Captcha.irctc vs. Python-based deep learning tools: which is better?

If you're proficient in Lua and Torch, this project offers high accuracy for IRCTC. For broader compatibility or familiarity, Python frameworks like TensorFlow might be easier to integrate, but may require custom tuning for similar performance.

Open-Awesome

captcha.irctc

Lua

A deep learning model that reads IRCTC captchas with 98% accuracy, demonstrating their vulnerability to automated booking.

GitHub

215 stars40 forks0 contributors

What is captcha.irctc?

Captcha.irctc is a deep learning project that trains neural networks to automatically read and solve captcha images from the IRCTC railway booking website. It demonstrates how machine learning can bypass traditional security measures designed to stop bots, achieving 98% accuracy in reading the text from these captchas. The project highlights the vulnerability of IRCTC's current captcha system to automated ticket booking software.

Target Audience

Machine learning researchers, security enthusiasts, and developers interested in computer vision, captcha vulnerabilities, or applying deep learning to real-world problems.

Value Proposition

It provides a working, high-accuracy model specifically for IRCTC captchas, using a residual network architecture and custom training criteria. The project serves as a practical case study in breaking captcha systems with deep learning, complete with code and methodology.

Overview

Reading irctc captchas with 98% accuracy using deep learning

Use Cases

Best For

Studying captcha vulnerability with deep learning
Training neural networks on sequential character recognition tasks
Building automated systems for IRCTC ticket booking
Learning Torch-based computer vision projects
Experimenting with residual networks for image classification
Understanding data augmentation for captcha datasets

Not Ideal For

Developers needing a plug-and-play captcha solver for multiple websites
Teams using modern deep learning frameworks like TensorFlow or PyTorch
Projects without access to GPU hardware for training
Applications requiring real-time captcha solving without model retraining

Pros & Cons

Pros

High Accuracy Performance

Achieves 98% accuracy on IRCTC captchas using a residual network architecture, as demonstrated in the README with test results.

Advanced Neural Architecture

Employs a 34-layer residual network and custom MultiCrossEntropyCriterion for sequential character recognition without RNNs, based on VGG designs.

Effective Data Augmentation

Implements augmentation techniques to improve model robustness, enhancing performance on varied captcha samples as noted in the code updates.

GPU-Optimized Training

Designed for NVIDIA GPUs like Titan 780, speeding up training significantly, as stated in the requirements section.

Cons

Specific to IRCTC Captchas

The model is trained only for IRCTC's captcha style and requires modifications for other websites, limiting generalizability.

Complex Setup Process

Requires installation of Torch, CUDA, cutorch, and csvigo, which can be challenging for users unfamiliar with these niche tools.

Large Dataset Dependency

Needs around 10,000 labeled samples for training, as admitted in the README, making it resource-intensive to replicate for new applications.

Ethical and Legal Risks

Demonstrates vulnerability that could be misused for automated ticket booking, raising concerns about responsible use and potential violations.

Frequently Asked Questions

Related Projects

buster

Captcha solver extension for humans, available for Chrome, Edge and Firefox

Stars9,211

Forks686

Last commit27 days ago

captcha_trainer

[验证码识别-训练] This project is based on CNN/ResNet/DenseNet+GRU/LSTM+CTC/CrossEntropy to realize verification code identification. This project is only for training the model.

Stars3,210

Forks823

Last commit8 months ago

captcha_break

验证码识别

Stars2,816

Forks668

Last commit4 years ago

uncaptcha

Defeating Google's audio reCaptcha with 85% accuracy.

Stars2,810

Forks327

Last commit8 years ago

Community-curated · Updated weekly · 100% open source

Found a gem we're missing?

Open-Awesome is built by the community, for the community. Submit a project, suggest an awesome list, or help improve the catalog on GitHub.

Submit a project Star on GitHub