File size: 3,750 Bytes
65bf862 eca797b 65bf862 eca797b | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 | ---
license: apache-2.0
tags:
- ocr
- rapidocr
- paddleocr
- mirror
library_name: rapidocr
---
# RapidOCR model mirror
A mirror of the [RapidAI/RapidOCR](https://www.modelscope.cn/models/RapidAI/RapidOCR)
model repository, re-hosted on Hugging Face so that downstream services can
fetch the weights from Hugging Face instead of `modelscope.cn`.
The files here are **byte-for-byte copies** of the corresponding files in the
upstream ModelScope repository at tag **`v3.9.2`** (302 files, ~3.6 GB: ONNX,
MNN, Paddle, and Torch model formats, character dictionaries, fonts, and test
files). No files were modified; relative paths are preserved.
## URL mapping
A file at the upstream revision URL
```
https://www.modelscope.cn/models/RapidAI/RapidOCR/resolve/v3.9.2/onnx/PP-OCRv6/rec/PP-OCRv6_rec_small.onnx
```
is mirrored at
```
https://huggingface.co/DjB314/RapidOCR/resolve/main/v3.9.2/onnx/PP-OCRv6/rec/PP-OCRv6_rec_small.onnx
```
In other words, the upstream revision prefix `resolve/v3.9.2/…` maps to
`resolve/main/v3.9.2/…` here, and the directory layout below the version
matches upstream exactly. Any tool that swaps the host + repo can therefore
use this mirror without changing paths.
## Usage
Direct download:
```bash
curl -L -o PP-OCRv6_rec_small.onnx \
https://huggingface.co/DjB314/RapidOCR/resolve/main/v3.9.2/onnx/PP-OCRv6/rec/PP-OCRv6_rec_small.onnx
```
Hugging Face Hub:
```bash
hf download DjB314/RapidOCR v3.9.2/onnx/PP-OCRv6/rec/PP-OCRv6_rec_small.onnx --local-dir .
# or the whole tree:
hf download DjB314/RapidOCR --repo-type model --local-dir rapidocr-models
```
Downstream tools that previously fetched from the upstream ModelScope URL
can switch to this mirror by swapping the host+repo in the URL — see the
[URL mapping](#url-mapping) section. The relative paths below the version are
identical to upstream, so no path changes are needed.
The files carry no modification timestamps from upstream, so integrity
checking is by checksum: see [verification](#verification).
## Versioning & refresh policy
| Revision mirrored | Files | Size | Date |
|---|---|---|---|
| `v3.9.2` | 302 | ~3.6 GB | 2026-08 |
This mirror tracks upstream **tags only** (never arbitrary `main` changes).
When RapidAI tags a new release, the mirror is re-synced and this table is
updated. If you depend on a newer tag before that happens, open an issue.
## Verification
Each staged file is copied verbatim from upstream and checked against the
SHA-256 recorded in the upstream ModelScope file listing; mismatches are
re-downloaded or reported, never shipped. To verify a downloaded file locally:
```bash
sha256sum PP-OCRv6_rec_small.onnx
# compare against the upstream listing:
# https://www.modelscope.cn/api/v1/models/RapidAI/RapidOCR/repo/files?Revision=v3.9.2&Recursive=true
```
## License & attribution
Licensed under the **Apache License, Version 2.0** — the same license declared
by the upstream sources. See [LICENSE](LICENSE).
- **RapidOCR** — [github.com/RapidAI/RapidOCR](https://github.com/RapidAI/RapidOCR)
and the model repository
[modelscope.cn/models/RapidAI/RapidOCR](https://www.modelscope.cn/models/RapidAI/RapidOCR)
(Apache-2.0). Original source of the files mirrored here.
- **PaddleOCR (PP-OCR)** — [github.com/PaddlePaddle/PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)
(Apache-2.0). Upstream origin of the PP-OCR model weights.
This mirror redistributes the above works **unmodified**, retaining their
Apache-2.0 license and attribution as required by Section 4 of the license.
This is an **independent mirror**. It is not affiliated with, endorsed by, or
sponsored by RapidAI, ModelScope, or PaddlePaddle. Product names and marks are
used for identification only.
|