---
license: apache-2.0
tags:
  - ocr
  - paddleocr
  - onnx
---

# Hako OCR model assets

Mirror of the OCR model assets bundled with the Hako application, organized by platform.
These are copies of upstream PaddleOCR assets, not models trained by hb-dev.
The model files were copied without modification from the application packages.

## Credits and license

PaddleOCR is developed by PaddlePaddle and its contributors:
https://github.com/PaddlePaddle/PaddleOCR

The ppu-paddle-ocr model distribution publishes ONNX/ORT conversions under Apache-2.0:
https://github.com/PT-Perkasa-Pilar-Utama/ppu-paddle-ocr-models
https://huggingface.co/snowfluke/ppu-paddle-ocr-models

The included LICENSE reproduces that distribution's Apache-2.0 license.
This mirror is maintained independently by hb-dev and does not imply upstream endorsement.

## Files

- `web/`: PP-OCRv5 mobile detection, recognition, and character dictionary used by the browser.
- `windows/`: PP-OCRv5 mobile detection, recognition, and character dictionary used by Windows.
- `android/det/`: the Android detector in ORT format.
- `android/rec/`: the Android recognizer in ORT format with its inference configuration.
  The configuration identifies the recognizer as `PP-OCRv6_small_rec`.

The web and Windows recognition files have the same filename but different SHA-256 values.
Keep them in their respective platform folders. The Android detector and recognizer also share
the filename `inference.ort`; preserve their `det/` and `rec/` folders.

`files-manifest.json` records byte counts, SHA-256 checksums, application source paths,
and any exact matches found in the inspected upstream conversion repository.

Uploading these files does not change the application's configured download locations.