Eng-Ahmd / break-captcha-with-capsule-networks

Breaking CAPTCHA with Capsule Networks

Folder structure

Data generation folders:

• captcha_generator: Generate "Dataset 1" mentioned in the paper.

• captcha_generator_split: Generate "Dataset 2" mentioned in the paper.

• cnn_captcha: The implementation of Deep-CAPTCHA model used to classify "Dataset 1".

• dynamic_captcha: The implementation of CapsNet model used to classify "Dataset 1".

• cnn_captcha_split: The implementation of Deep-CAPTCHA model used to classify "Dataset 2".

• dynamic_captcha_split: The implementation of Deep-CAPTCHA model used to classify "Dataset 2".

Experiments

4.1 Adapt the CapsNet to the CAPTCHA dataset

1. Run python captcha_generator.py inside folder captcha_generator to generate "CAPTCHA_4digits_noise" (Dataset 1)
2. Copy the folder "CAPTCHA_4digits_noise" to cnn_captcha/data or dynamic_captcha/data as the dataset for those two models
3. Configure the script in cnn_captcha/local_experiment_scripts or dynamic_caps_captcha/local_experiment_scripts
4. Use bash cnn_captcha4digit_test.sh or bash dynamic_captcha4digit_test.sh to run training jobs. The results will be stored in a folder with the name specified in the script.

4.2 Adapt CapsNet to the CAPTCHA Puzzle Task

1. Run python captcha_generator.py inside folder captcha_generator_split to generate "CAPTCHA_3digits_noise" (Dataset 2)
2. Copy the folder "CAPTCHA_3digits_noise" to cnn_captcha_split/data or dynamic_caps_captcha_split/data as the dataset for those two models
3. Configure the script in cnn_captcha_split/local_experiment_scripts or dynamic_caps_captcha_split/local_experiment_scripts
4. Use bash cnn_captcha4digit_test.sh or bash dynamic_captcha4digit_test.sh to run training jobs. The results will be stored in a folder with the name specified in the script.