PaddleOCR2Pytorch
PaddleOCR2Pytorch is a PyTorch port of PaddleOCR, enabling models trained with PaddleOCR to run natively within the PyTorch framework. The project provides a reference implementation for converting PaddlePaddle models to PyTorch while supporting a comprehensive suite of OCR tasks. Key features include multiple text detection algorithms such as DB, DB++, SAST, EAST, PSENet, and FCENet, along with text recognition models including CRNN, RARE, Rosetta, STARNet, SRN, NRTR, SAR, SVTR, ViTSTR, and PGNet. It supports the PP-OCR series (v1 through v5) with both mobile and server variants, offering high accuracy across Chinese, English, and multilingual scenarios including Korean, Japanese, French, German, and Arabic among over 27 supported languages. Additional capabilities include formula recognition via the CAN model, text super-resolution through Text Telescope, support for handwritten text, vertical text, and long text recognition. PP-OCRv5 introduces a single model handling five text types: Simplified Chinese, T