The PaddlePaddle team at Baidu has unveiled the latest iteration of PaddleOCR, version 3.0, marking significant advancements in text recognition technology. This new release showcases enhancements across various fronts, including heightened accuracy, expanded multilingual support, improved handwritten text recognition, and enhanced document parsing capabilities. Furthermore, PaddleOCR 3.0 is fully compatible with the PaddlePaddle Framework 3.0 and now supports domestic hardware, broadening its applicability. The standout feature, the PP-OCRv5 model, boasts support for five text types and complex scenarios, while delivering a notable 13% increase in accuracy. Additionally, PP-StructureV3 upgrades document parsing capabilities, and PP-ChatOCRv4, integrated with larger models, achieves a 15% improvement in the accuracy of key information extraction.
