Implements end-to-end Chinese OCR with two stages: CTPN (Connectionist Text Proposal Network) for detecting text regions and a DenseNet encoder with CTC (Connectionist Temporal Classification) decoder for sequence recognition. Built on Tensorflow and Keras with complete training scripts, synthetic data generation support, and demo evaluation pipeline. The model achieves 98.3% validation accuracy at 8ms inference time.