README.md

---
license: apache-2.0
library_name: onnx
tags:
  - ocr
  - handwriting-recognition
  - paddleocr
  - onnx
  - multilingual
language:
  - en
  - zh
  - ja
  - hi
  - th
  - ar
  - ko
---

# DocScanner on-device OCR models

ONNX exports of [PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR) text detection and
recognition models, used by [DocScanner](https://github.com/Bible-Translation-Tools/doc-scanner)
to transcribe handwritten pages **entirely on the device**.

DocScanner is a free, open-source tool for Bible translation field teams. Transcription has to
work without a network and without per-page cost, so no cloud model is involved: a page is cut
into text lines by the detector, and each line is read by the recognizer for that project's
writing system.

## Layout

A device downloads the detector plus exactly one recognizer — 13 MB for most scripts, 78 MB for
Latin/CJK — never the whole set.

| path | file | size | sha256 (short) |
|------|------|------|----------------|
| `detector/` | `det_model.onnx` | 4.6 MB | `0c5eeee2` |
| `ppocrv6/` | `rec_model.onnx` | 73 MB | `4078550d` |
| `ppocrv6/` | `charset.json` | 128 KB | `46f80089` |
| `devanagari/` | `rec_model.onnx` | 7.6 MB | `a3d5b5fa` |
| `devanagari/` | `charset.json` | 3.6 KB | `621d0073` |
| `thai/` | `rec_model.onnx` | 7.5 MB | `d52231fe` |
| `thai/` | `charset.json` | 3.3 KB | `b62ede69` |
| `arabic/` | `rec_model.onnx` | 7.6 MB | `a1e69c68` |
| `arabic/` | `charset.json` | 4.5 KB | `df730929` |
| `korean/` | `rec_model.onnx` | 13 MB | `03525a1d` |
| `korean/` | `charset.json` | 81 KB | `e600744e` |

The app pins a **tag** of this repository in its download URL, so a given build can only ever
fetch the weights it was tested against.

## Provenance

Exported with [`paddle2onnx`](https://github.com/PaddlePaddle/Paddle2ONNX) from the official
PaddleOCR inference models, unquantized (float32):

| directory | source model | opset |
|-----------|--------------|-------|
| `detector/` | `PP-OCRv5_mobile_det` | 17 |
| `ppocrv6/` | `PP-OCRv6_medium_rec` | 16 |
| `devanagari/` | `devanagari_PP-OCRv5_mobile_rec` | 16 |
| `thai/` | `th_PP-OCRv5_mobile_rec` | 16 |
| `arabic/` | `arabic_PP-OCRv5_mobile_rec` | 16 |
| `korean/` | `korean_PP-OCRv5_mobile_rec` | 16 |

`charset.json` is each model's character table from PaddleOCR, as a JSON array. The recognizer
output has `charset + 2` classes: index 0 is the CTC blank, then the table, then a space —
the layout PaddleOCR's `CTCLabelDecode` expects.

## Interfaces

**Detector** (`det_model.onnx`) — input `[1, 3, H, W]`, BGR, long side scaled to 960 and floored
to a multiple of 32, normalized with ImageNet statistics (mean `0.485/0.456/0.406`, std
`0.229/0.224/0.225`). Output `[1, 1, H, W]`, a per-pixel probability of text.

**Recognizers** (`rec_model.onnx`) — input `[1, 3, 48, W]`, RGB, height 48 with width scaled by
the line's aspect ratio and padded to a multiple of 32, normalized `(x/255 - 0.5) / 0.5`.
Padding is left at zero *after* normalization (mid-grey), which is how PaddleOCR pads; padding
with black instead wrecks recognition. Output `[1, T, classes]` logits, greedy CTC decoded.

Arabic is read right-to-left: CTC scans left to right and so emits the logically last character
first. The line is reversed by grapheme cluster, keeping combining marks attached to their base
letter — correcting this took the error on rendered Arabic from 79% to 18%.

## Measured accuracy

Character error rate on handwritten sample pages, per line, after detection:

| script | CER | note |
|--------|-----|------|
| English (neat) | 5-7% | |
| English (hard hand) | ~14% | |
| Spanish / French | ~4% | |
| Chinese | 2% | written in a squared-paper grid |
| Devanagari | usable | whole lines, most words legible |
| Thai | 15-25% | |
| Arabic | 25-35% | words land in the right places and reading order |
| Korean | 15-25% | several lines nearly verbatim |

These are CTC models with **no language model**, which is deliberate: an unreadable crop comes
back garbled or empty rather than as fluent invented text. A TrOCR alternative scored better on
neat English (2%) but rewrote what it could not read — including inventing liturgical Church
Slavonic from a blank strip — which is the wrong failure mode for a translation tool.

## Not included

- **Cyrillic** — PaddleOCR's Cyrillic recognizers manage only 76-84% CER on handwriting, so
  those projects use server transcription until a model trained on handwritten lines exists.
- **Hebrew** — no PaddleOCR model.
- **Greek** — not yet wired.

## License

Apache-2.0, inherited from PaddleOCR. Please keep the attribution to the PaddleOCR project.