PaddleOCR v2.8.0, as a milestone update of the text recognition development kit under the PaddlePaddle deep learning open-source framework, introduces cutting-edge OCR technology. This version includes the winning solutions from the PaddleOCR Algorithm Model Challenge, such as the Scene Text Recognition algorithm SVTRv2 and the Table Recognition algorithm SLANet-LCNetV2, setting new standards for the OCR field.
At the same time, the project structure has been deeply optimized, with non-core modules migrated to new repositories, allowing the project to focus more on OCR core technology. In addition, historical difficult problems such as the model not running after updating Backbone, numpy version dependency conflicts, and slow performance on the Mac system have been resolved, enhancing user experience.
The new version also includes fixes for issues such as the loss of OCR results in layout analysis, the introduction of pyproject.toml to comply with PEP518 standards, and optimization improvements such as the sliding window operation for large image inference, enhancing the software's stability, compatibility, and performance. The support and contributions from the open-source community are crucial for every progress of PaddleOCR v2.8.0, and the efforts of PMC members and contributors are particularly thanked.
PaddleOCR is building a dedicated documentation tutorial site that will provide keyword search functionality and an elegant and comfortable interface.
Project Address: https://github.com/PaddlePaddle/PaddleOCR