Third-party model files bundled with OCR Lab
===========================================

This directory contains machine-learning model weights that OCR Lab **redistributes
unmodified**. They are not the work of OCR Lab.

## PP-OCRv6 (tiny)

- **Copyright** © PaddlePaddle Authors.
- **Project:** PaddleOCR — https://github.com/PaddlePaddle/PaddleOCR
- **Licence:** Apache License 2.0. The full text is in `LICENSE-Apache-2.0.txt` in this
  directory.
- **Modifications:** none. The `.onnx` and `.yml` files are byte-for-byte the published
  artifacts; the SHA-256 of each is recorded below and can be checked against the source.

| File here | Source repository | Revision | Bytes | SHA-256 |
| --- | --- | --- | --- | --- |
| `PP-OCRv6_tiny_det.onnx` | `huggingface.co/PaddlePaddle/PP-OCRv6_tiny_det_onnx` | `2ba1506c0380b8f0b03dd142459aac66d4421f6c` | 1,780,590 | `193bab7a04fca699a6c82e6abb5b81bdb28177f0abd4062552b04908dafb19f8` |
| `PP-OCRv6_tiny_rec.onnx` | `huggingface.co/PaddlePaddle/PP-OCRv6_tiny_rec_onnx` | `2612ab37152ae0a677521bae4e1e3d4fb4cf7c30` | 4,462,639 | `9ef676d6ed3c88256a2d92c640c44f25b0c40947e111b14b8be8f594091563e6` |
| `PP-OCRv6_tiny_rec.yml` | `huggingface.co/PaddlePaddle/PP-OCRv6_tiny_rec_onnx` | `2612ab37152ae0a677521bae4e1e3d4fb4cf7c30` | 55,571 | `66170210bad538e83fff3c4a3867e547d6bf20b50d64b20347c4b913f3034ea1` |

`PP-OCRv6_tiny_rec.yml` is the recognition package's `inference.yml`. It carries the
6,904-entry `PostProcess.character_dict` that the CTC decoder maps class indices through,
plus the preprocessing constants the pipeline reads at load time instead of hardcoding.

Each file was reviewed through the platform's vetting lane before being bundled: only
default `ai.onnx` operators, no custom operator domains, no local functions, and zero
initializers referencing external data.

## onnxruntime-web

The runtime files at the bundle root (`ort.wasm.min.js`, `ort-wasm-simd-threaded.mjs`,
`ort-wasm-simd-threaded.wasm`) are onnxruntime-web 1.27.0, © Microsoft Corporation, MIT
licence — https://github.com/microsoft/onnxruntime. Also shipped unmodified.

Demo image — sample.png
-----------------------

`sample.png` is a screenshot of LibreOffice Impress 3.3 by The Document Foundation,
released under the Creative Commons CC0 1.0 Universal Public Domain Dedication and
shipped here byte-for-byte unmodified (sha256
5a8e67f064811819888d7165bfac828e00a25f45b4139b63d449521594bb9d25, 1365x721,
112,197 bytes). CC0 imposes no conditions; the credit is given because it is deserved,
not because it is required.

  Source: https://commons.wikimedia.org/wiki/File:LibreOffice_Impress_3.3.png

It is used as the default demo because a real application screenshot is a far more
honest test than clean prose: about half its recognised lines fall below the 0.75
confidence mark, since toolbar glyphs genuinely do read as one-character noise.

`sample-zh.png` is a synthetic Chinese/English test image generated for this app. It is
not third-party content.