PP-OCR language packs
Recognition models that OverTranslate downloads the first time someone chooses a language that is not in its installer. Each folder holds one writing system: a recogniser and its dictionary. The app shares its detector and classifier across all of them.
| Folder | Reads | Files | Upstream model |
|---|---|---|---|
cyrillic |
Russian, Ukrainian (the dictionary covers the Cyrillic alphabet) | cyrillic-v1-rec.onnx, cyrillic-v1-dict.txt |
cyrillic_PP-OCRv5_rec_mobile, ppocrv5_cyrillic_dict.txt |
korean |
Korean | korean-v1-rec.onnx, korean-v1-dict.txt |
korean_PP-OCRv5_rec_mobile, ppocrv5_korean_dict.txt |
Versions
The version is in the file name. A new version is added as new files, and published files are never replaced, because released builds of the app check them against a fixed SHA-256.
| File | Bytes | SHA-256 |
|---|---|---|
cyrillic/cyrillic-v1-rec.onnx |
8074092 | 90f761b4bfcce0c8c561c0cb5c887b0971d3ec01c32164bdf7374a35b0982711 |
cyrillic/cyrillic-v1-dict.txt |
2781 | db40aa52ceb112055be80c694afdf655d5d2c4f7873704524cc16a447ca913ba |
korean/korean-v1-rec.onnx |
13488748 | cd6e2ea50f6943ca7271eb8c56a877a5a90720b7047fe9c41a2e541a25773c9b |
korean/korean-v1-dict.txt |
47451 | a88071c68c01707489baa79ebe0405b7beb5cca229f4fc94cc3ef992328802d7 |
Source and license
The models are the work of the PaddleOCR authors. The ONNX conversions and dictionaries are unmodified copies from RapidAI/RapidOCR revision v3.9.2; only the file names were changed. PaddleOCR and RapidOCR are both released under the Apache License 2.0, included here as LICENSE.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support