RapidOCR provides the text detection and recognition models of PaddleOCR
converted to ONNX, so they can be run without the PaddlePaddle framework.
ONNX Runtime, OpenVINO, MNN, PaddlePaddle and PyTorch are supported as
inference engines.

Detection locates the text boxes in an image, recognition reads the text
inside them.  Models for Chinese, English, Japanese, Korean and further
languages are fetched on demand.