PaddleOCR ONNX Backup · PP-OCRv6 Text Recognition and PP-DocLayoutV3 Layout Detection
English | 简体中文 · Also on ModelScope · 42ailab.com
Two sets of small models that run on an ordinary CPU:
- PP-OCRv6 (small and medium): finds where the text is in an image (detection), then reads it (recognition).
- PP-DocLayoutV3: splits a document page into regions such as headings, body text, tables, figures and formulas, so the recognizer knows what to read and in which order.
The models were built by the Baidu PaddlePaddle PaddleOCR team and released under Apache-2.0. This repository is neither a new model nor our conversion. The ONNX files were converted from the official PaddlePaddle weights by GreatV, author of oar-ocr, and published in the personal ModelScope repository
greatv/oar-ocr. This repository is a byte-for-byte backup of six of those files.
Why a backup
These six files had a single source: one personal repository on ModelScope. If it is renamed, deleted or reorganized, people who depend on it lose their download.
So we keep the original files, unmodified, on ModelScope and on our own Hugging Face account. File names, sizes and sha256 digests all match the original.
Use
Packaged for 42model: download PP-OCRv6 from the OCR section of the desktop app, or run 42model download pp-ocrv6:small.
Files and license
| File | Role | Size | sha256 |
|---|---|---|---|
pp-ocrv6_small_det.onnx |
text detection (small) | 9.4 MiB | d73e0058b7a8086bbd57f3d10b8bcd4ff95363f67e06e2762b5e814fe9c9410e |
pp-ocrv6_small_rec.onnx |
text recognition (small) | 20.2 MiB | 5435fd747c9e0efe15a96d0b378d5bd157e9492ed8fd80edf08f30d02fa24634 |
pp-ocrv6_medium_det.onnx |
text detection (medium) | 59.2 MiB | eb13b44b25bb36f89528b68720af8a61d9cf381176107f465db1757b65d086e1 |
pp-ocrv6_medium_rec.onnx |
text recognition (medium) | 73.0 MiB | 9c09abf0957f7968c7586464b7397b84ad2387a0497a351af40e9acc71b673ba |
ppocrv6_dict.txt |
recognition dictionary (shared) | 73 KiB | b5f2bfe2bdd9448429e3e82b51c789775d9b42f2403d082b00662eb77e401c5d |
pp-doclayoutv3.onnx |
layout detection | 123.9 MiB | 1a7ec3812d239ad14debb87e38273012558fea27ac10b44b61260e3a88358e39 |
The digests can be checked one by one against the file pages of the original greatv/oar-ocr repository.
Weights are under the upstream Apache-2.0 license; the full text is in this repository's LICENSE (taken from the PaddleOCR repository, copyright PaddlePaddle Authors). Neither the PaddleOCR repository nor the official PaddlePaddle model repositories ship a separate NOTICE file. Credit for the ONNX conversion goes to GreatV (oar-ocr).