PP-OCRv5 mobile ONNX models bundled with kebab (paddle-onnx OCR engine)
=======================================================================

These model weights and the recognition dictionary are derived from
PaddleOCR (https://github.com/PaddlePaddle/PaddleOCR), licensed under the
Apache License, Version 2.0.

  Copyright (c) PaddlePaddle Authors.
  Licensed under the Apache License, Version 2.0 (the "License");
  you may not use these files except in compliance with the License.
  You may obtain a copy of the License at
      http://www.apache.org/licenses/LICENSE-2.0

Files
-----
  ppocrv5_mobile_det.onnx          PP-OCRv5_mobile detection model (DBNet)
  korean_ppocrv5_mobile_rec.onnx   korean_PP-OCRv5_mobile recognition model (CTC)
  korean_dict.txt                  recognition dictionary (11,945 chars: KR + Latin + digits + symbols)

These were converted from the official PaddlePaddle inference models to ONNX
via paddle2onnx for in-process execution with onnxruntime (`ort`). No model
architecture or weights were modified; only the serialization format changed.

The recognition CTC class layout (empirically confirmed, see
tests/golden/ctc_rec_golden.json):
  index 0          = CTC blank
  index 1..11945   = korean_dict.txt line N -> class N (dict[N-1])
  index 11946      = space ' '
  total classes    = 11947 (= 11945 dict + blank + space)

If any post-processing source (min-area-rect / polygon unclip) is later
ported verbatim from oar-ocr (Apache-2.0), record the per-file provenance
here as required by the Apache-2.0 attribution clause.
