This project demonstrates how to use Keras and TensorFlow to construct a deep convolutional neural network for captcha recognition. It uses CRNN architecture with CTC (Connectionist Temporal Classification) loss for sequence prediction, supporting both TensorFlow and PyTorch implementations. The notebook includes code for generating captcha training data, visualizing model architecture, and training models efficiently with GPU acceleration including CuDNNGRU layers.