Radar ·
MiniCPM-OCR suite: open OCR models and an accuracy verifier
OpenBMB announces two open-source models for OCR: one for document parsing and another that checks the accuracy of text, formulas and tables extracted by different parsers.
According to the source, OpenBMB announced two open-source models. MiniCPM-OCR 1.0 performs document parsing at the 1B parameter scale. MiniCPM-OCR Jev, with 2B parameters, verifies the accuracy of text, formula and table parsing produced by different OCR parsers. The author reports benchmark scores and performance claims, but the available text is truncated, so the figures are not reproduced here.
The relevance lies in the verification role: a model that reviews the output of other extractors can help check document extraction results. Rota Nacional does not offer this verifier and does not claim to solve the OCR accuracy problem described. If your team uses these models to study or test extraction, do not send documents containing personal data without prior anonymization, and check results against the original. To consult the source, use the project's official record and read the license and full benchmarks before any use.
Original source: Suíte MiniCPM-OCR: modelos de OCR de código aberto e verificador de precisão