This repository implements SimGAN to solve captchas without requiring manual label collection. It uses a generative adversarial network to refine synthetic captcha images so they resemble real ones, then trains a classifier on the refined synthetic data. The approach leverages Keras for the neural network implementation.