MinerU · Surya · Docling
Compare document parsing, layout order, tables, and format compatibility; Chinese quality and hardware needs still require testing.
Decision guide · OCR
How should you choose local Chinese OCR?
If the goal is to turn scanned PDFs, papers, or tables into searchable, processable text, compare more than recognition rate: layout, tables, deployment, licensing, and whether the evidence supports a decision.
Compare these three layers
- Input and output Scanned PDFs, images, tables, Markdown, or structured fields?
- Deployment and cost Offline local, GPU, cloud API, and data-compliance requirements?
- Verifiability What does the KG know, and what still needs a test on real documents?
A candidate is not an answer
Do not trust type filters alone
OCR candidates may be recorded as models, tools, or document-processing entities. Search the complete need to retrieve them together.