occular-ocr

Occular-OCR — model weights

Model weights for Occular-OCR: a text detector, text recognizers (svtr_lcnet and svtr_t, for Russian/English and for 12 Cyrillic-script languages), a page-orientation model, per-language language models, and an optional reading-order model. The code downloads these files on first use.

License

Most weights are NOT under a permissive license. The detector, the recognizers (svtr_lcnet, svtr_t), the language models, and the optional reading-order model are licensed under a modified AI Pubs OpenRAIL-M license — see LICENSE.md.

• Free for individuals, researchers, non-profits, and small organizations (annual revenue under 20 000 000 ₽ and fewer than 8 employees).
• Larger organizations need a commercial license — 300 000 ₽ / year per organization.
• Standard OpenRAIL-M use restrictions apply (no disinformation, no unlawful or rights-violating use, no unlawful discrimination), and must be passed on to downstream users.

Exception — orientation_orinet_fp32.onnx (page-orientation model) is licensed under Apache-2.0, not OpenRAIL-M. That file is free for any use, including commercial, subject only to attribution; the restrictions and commercial terms above do not apply to it.

The Occular-OCR source code is separately licensed under Apache-2.0; this repository's license governs only the weight files here.

Commercial licensing: [email protected] · Telegram @Bodhi_b

Attribution

If you distribute these weights or a fine-tuned derivative, retain this license and credit Occular-OCR.