captcha_trainer
captchatrainer is a deep learning based CAPTCHA recognition training tool built on TensorFlow 1.14. It uses architectures like CNN5, ResNet50, and DenseNet combined with GRU, LSTM, or BiLSTM recurrent layers, and supports both CTC and CrossEntropy loss functions. The network extracts features through convolutional layers, predicts label distributions through recurrent layers, and converts predictions into final results through a transcription layer. The tool is designed to handle complex CAPTCHA scenarios including overlapping characters, perspective distortion, blur, and noise interference. It features a project-based management system where each image classification task gets its own independent workspace, allowing multiple models to coexist and be managed separately without modifying code. Key features include a pre-compiled Windows GPU version requiring no environment setup, incremental dataset additions without repackaging, flexible network configurations including CNN-only mode without recurrent layers,