Change OCR model ?

[GnosisLabsG]

MinerU using PaddleOCR, but am i able to change the OCR ? also why dont they add multi OCR support so we can use other OCRs like LightOnOCR-2-1B

[dosubot[bot]]

Hi @GnosisLabsG! I'm Dosu and I’m helping the MinerU team.

MinerU only supports PaddleOCR as its OCR engine—there’s no built-in way to switch to a different OCR engine like LightOnOCR-2-1B or to use multiple OCR engines at once. All CLI/API options (like --lang or --source) only affect which PaddleOCR model is used, not the underlying engine itself.

If you want to use a different OCR engine, you’d need to modify the codebase, specifically the OCR model initialization logic (e.g., ocr_model_init and related classes), to instantiate your own OCR implementation that matches the expected interface and output format. There’s no abstraction layer or plugin system for swapping OCR engines, so this isn’t a plug-and-play change—it requires development work and a good understanding of the codebase source.

There is a limited extensibility point: advanced users can process the img_crop_list with their own OCR engine by modifying batch_analyze.py, but this is not officially documented or supported source.

There are no current plans or roadmap for multi-OCR support or for supporting engines like LightOnOCR-2-1B. Adding this would require significant architectural changes, such as introducing an abstraction/interface for OCR engines and configuration options to select between them.

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}