How paddlepaddle/paddleocr works
PaddleOCR is a broad OCR and document-AI toolkit: users submit images or PDFs through its Python interfaces, run text recognition or document-parsing pipelines, and receive structured results such as JSON or Markdown. The repository also includes model/data-training infrastructure, API clients, and deployment runtimes. The graph emphasizes these product capabilities; some pipeline links are documented at the product level because their implementation was not sampled, so uncertain connections are left dashed.
Main entry points: __init__.py Python API · base.py Inference models · client.py API SDK clients · pp_structurev3.py Structure parsing
Components 22
Python inference
- Python CLI
paddleocr/_cli.py - Python API
paddleocr/__init__.py - OCR pipeline
paddleocr/_pipelines/ocr.py - Document preprocessing
paddleocr/_pipelines/doc_preprocessor.py - Structure parsing
paddleocr/_pipelines/pp_structurev3.py - Document VLM
paddleocr/_pipelines/paddleocr_vl.py - Document conversion
paddleocr/_doc2md/core.py
Models and training
- Inference models
paddleocr/_models/base.py - Model architectures
ppocr/modeling/architectures/base_model.py - Training runner
tools/program.py - Training data
ppocr/data/simple_dataset.py
API clients
- API SDK clients
paddleocr/_api_client/client.py - Async SDK
paddleocr/_api_client/async_client.py - HTTP transport
paddleocr/_api_client/_http.py - Job polling
paddleocr/_api_client/_poller.py - Result resources
paddleocr/_api_client/_resources.py
Deployment and integrations
- C++ inference
deploy/cpp_infer/src/pipelines/ocr/pipeline.cc - Mobile OCR engines
deploy/ppocr-android/ppocr-sdk/src/main/java/com/paddle/ocr/engine/OCREngine.kt - MCP integration
mcp_server/paddleocr_mcp/tasks/ocr.py - LangChain loader
langchain-paddleocr/langchain_paddleocr/document_loaders/paddleocr_vl.py
Other components
- User or application
- PaddleOCR API service
Connections 26
- User or application → Python CLI: invokes
- User or application → Python API: submits images
- Python CLI → Python API: dispatches
- Python API → OCR pipeline: runs OCR
- Python API → Document preprocessing: runs preprocessing
- Python API → Structure parsing: parses structure
- Python API → Document VLM: parses with VLM
- OCR pipeline → Inference models: uses inference
- Document preprocessing → Inference models: uses inference
- Structure parsing → Inference models: uses inference
- Document VLM → Inference models: uses inference
- Structure parsing → Document conversion: produces documents
- Document VLM → Document conversion: produces documents
- Training data → Training runner: supplies batches
- Training runner → Model architectures: trains models
- API SDK clients → HTTP transport: uses transport
paddleocr/_api_client/client.py - API SDK clients → Job polling: polls jobs
paddleocr/_api_client/client.py - API SDK clients → Result resources: saves results
paddleocr/_api_client/client.py - Async SDK → HTTP transport: uses transport
paddleocr/_api_client/async_client.py - Async SDK → Job polling: polls jobs
paddleocr/_api_client/async_client.py - HTTP transport → PaddleOCR API service: submits jobs
- PaddleOCR API service → Job polling: returns job state
- User or application → C++ inference: runs locally
- User or application → Mobile OCR engines: runs on device
- User or application → MCP integration: connects agent
- User or application → LangChain loader: loads documents