D4ichi PassGAN Save

PassGAN source code for Python 3 & TensorFlow 1.13 with a pre-trained model. https://arxiv.org/abs/1709.00440

Project README

PassGAN

This repository is updated version of @brannondorsey/PassGAN for Python 3 & TensorFlow 1.13, contains code for the PassGAN: A Deep Learning Approach for Password Guessing paper.

The model from PassGAN is taken from Improved Training of Wasserstein GANs and it is assumed that the authors of PassGAN used the improved_wgan_training tensorflow implementation in their work.

This repo contributes:

  • A command-line interface sample.py train.py
  • A pretrained PassGAN model trained on the RockYou dataset
  • Jupyter notebook for debugging notebook-sample.py notebook-train.py

Getting Started

# requires CUDA 8 to be pre-installed
pip3 install -r requirements.txt

Generating password samples

Use the pretrained model to generate 1,000,000 passwords, saving them to generated_pass.txt.

python sample.py \
	--input-dir pretrained \
	--checkpoint pretrained/checkpoints/checkpoint_200000.ckpt \
	--output generated_pass.txt \
	--batch-size 1024 \
	--num-samples 1000000

Training your own models

You can downlaod sample datasets from release page, or generate sample rockyou dataset by yourself with codes under bin.

Training a model on a large dataset (100MB+) can take several hours on a GTX 1080.

# download the rockyou training data
# contains 80% of the full rockyou passwords (with repeats)
# that are 10 characters or less
curl -L -o data/train.txt https://github.com/d4ichi/PassGAN/releases/download/data/rockyou-test.txt

# train for 200000 iterations, saving checkpoints every 5000
# uses the default hyperparameters from the paper
python train.py --output-dir output --training-data data/train.txt

You are encouraged to train using your own password leaks and datasets. Some great places to find those include:

Attribution and License

This code is released under an MIT License. You are free to use, modify, distribute, or sell it under those terms.

The credit for the code in this repository goes to @igul222 for his work on the improved_wgan_training and @brannondorsey for specializing it in the PassGAN paper.

This is updated version for Python 3 / TensorFlow 1.13 of their work.

The PassGAN research and paper was published by Briland Hitaj, Paolo Gasti, Giuseppe Ateniese, Fernando Perez-Cruz.

Open Source Agenda is not affiliated with "D4ichi PassGAN" Project. README Source: d4ichi/PassGAN
Stars
127
Open Issues
9
Last Commit
1 year ago
Repository
License
MIT

Open Source Agenda Badge

Open Source Agenda Rating