BaoBao.PaddleOCR.Models 0.1.0

BaoBaoPaddleOCR.NET

BaoBaoPaddleOCR.NET is a local OCR wrapper built as C++ inference + .NET P/Invoke.

• BaoBaoPaddleOCR.Native: native C ABI layer
• BaoBaoPaddleOCR: C# wrapper
• BaoBaoPaddleOCR.Cli: CLI for smoke tests and local debugging

Current defaults

The repo is now wired for the assumptions we have so far:

• PaddleOCR source root: bundled in this repo at src/BaoBaoPaddleOCR.NET/PaddleOCR-3.4.0
• OpenCV root: C:\opencv\build
• Inference target: CPU
• Publish target: win-x64
• Model layout: defaults to official PaddleOCR inference directories under the model root:
  • PP-OCRv5_server_det_infer
  • PP-OCRv5_server_rec_infer
  • PP-LCNet_x1_0_textline_ori_infer

Local prerequisites

This repo already bundles:

• PaddleOCR-3.4.0
• paddle_inference

The machine still needs:

• Visual Studio C++ toolchain
• CMake or Visual Studio CMake tools
• OpenCV at C:\opencv\build or passed with -OpenCvDir

Build .NET

dotnet build .\src\BaoBaoPaddleOCR.NET\BaoBaoPaddleOCR.slnx

Pack NuGet

.\src\BaoBaoPaddleOCR.NET\pack-nuget.ps1 `
  -Version 0.1.0 `
  -Configuration Release `
  -BuildNative `
  -WithPaddleOcr `
  -WithModels `
  -PaddleInferenceDir .\src\BaoBaoPaddleOCR.NET\paddle_inference

Default output:

src/BaoBaoPaddleOCR.NET/artifacts/nuget/

Produced packages:

• BaoBao.PaddleOCR: managed wrapper DLL and native runtime assets under runtimes/win-x64/native
• BaoBao.PaddleOCR.Models: OCR models under contentFiles/any/any/models

By default, the models package copies bundled models to $(TargetDir)\models. Set BaoBaoPaddleOCRCopyModels=false in the consuming project if you want to manage models yourself.

Publish To Local Feed

.\src\BaoBaoPaddleOCR.NET\publish-local-feed.ps1

Default local feed:

src/BaoBaoPaddleOCR.NET/artifacts/local-feed/

Publish To NuGet.org

.\src\BaoBaoPaddleOCR.NET\publish-nugetorg.ps1 `
  -ApiKey "<your-nuget-api-key>" `
  -SkipDuplicate

Consumer Sample

Sample project:

src/BaoBaoPaddleOCR.NET/samples/NuGetConsumer

After packing and publishing to the local feed:

dotnet build .\src\BaoBaoPaddleOCR.NET\samples\NuGetConsumer\NuGetConsumer.csproj -c Release
dotnet run --project .\src\BaoBaoPaddleOCR.NET\samples\NuGetConsumer\NuGetConsumer.csproj -c Release -- .\src\BaoBaoPaddleOCR.NET\PaddleOCR-3.4.0\test_tipc\web\test.jpg

Build native mock DLL

.\src\BaoBaoPaddleOCR.NET\build-native.ps1 -Configuration Release

Build native PaddleOCR DLL

.\src\BaoBaoPaddleOCR.NET\build-native.ps1 `
  -Configuration Release `
  -WithPaddleOcr `
  -PaddleInferenceDir .\src\BaoBaoPaddleOCR.NET\paddle_inference

If you need GPU later:

.\src\BaoBaoPaddleOCR.NET\build-native.ps1 `
  -Configuration Release `
  -WithPaddleOcr `
  -WithGpu `
  -PaddleInferenceDir "D:\deps\paddle_inference" `
  -CudaLibDir "C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.6\lib\x64" `
  -CudnnLibDir "C:\tools\cudnn\lib\x64"

Run CLI

Mock mode:

$env:BAOBAO_PADDLEOCR_MOCK_TEXT = "mock result"
dotnet run --project .\src\BaoBaoPaddleOCR.NET\BaoBaoPaddleOCR.Cli\BaoBaoPaddleOCR.Cli.csproj -- .\demo.png --full

Real mode:

dotnet run --project .\src\BaoBaoPaddleOCR.NET\BaoBaoPaddleOCR.Cli\BaoBaoPaddleOCR.Cli.csproj -- .\demo.png --model-root .\models --full

Model layout

Default model root:

• CLI argument: --model-root <dir>
• or env: BAOBAO_PADDLEOCR_MODEL_ROOT
• or fallback: AppContext.BaseDirectory\models

Default expected layout:

models/
  PP-OCRv5_server_det_infer/
  PP-OCRv5_server_rec_infer/
  PP-LCNet_x1_0_textline_ori_infer/

If your actual subdirectory names are different, you can override them with environment variables:

• BAOBAO_PADDLEOCR_DET_DIRNAME
• BAOBAO_PADDLEOCR_REC_DIRNAME
• BAOBAO_PADDLEOCR_CLS_DIRNAME

These can be either:

• a child directory name under the model root, or
• an absolute path

Runtime environment variables

• BAOBAO_PADDLEOCR_NATIVE_DIR: extra directory to search for the native DLL
• BAOBAO_PADDLEOCR_MODEL_ROOT: model root directory
• BAOBAO_PADDLEOCR_MOCK_TEXT: mock output text when using the stub build
• BAOBAO_PADDLEOCR_CPU_THREADS: CPU thread count for PaddleOCR
• BAOBAO_PADDLEOCR_ENABLE_MKLDNN: true or false

Notes

• The native wrapper now calls PaddleOCR deploy/cpp_infer directly when built with -WithPaddleOcr.
• build-native.ps1 now auto-detects cmake.exe, Visual Studio developer tools, and falls back to Ninja when Visual Studio 17 2022 is unavailable.
• The build script copies BaoBaoPaddleOCR.Native.dll and common Paddle/OpenCV runtime DLLs into both:
  • BaoBaoPaddleOCR.Cli/bin/<Configuration>/net10.0/native
  • BaoBaoPaddleOCR/runtimes/win-x64/native
• The machine must still have a working MSVC/CMake toolchain installed.

Chinese quick start

# 1. 编译 .NET
dotnet build .\src\BaoBaoPaddleOCR.NET\BaoBaoPaddleOCR.slnx

# 2. 编译 stub native（不接 PaddleOCR，验证 P/Invoke 链路）
.\src\BaoBaoPaddleOCR.NET\build-native.ps1 -Configuration Release

# 3. 编译真实 PaddleOCR native
.\src\BaoBaoPaddleOCR.NET\build-native.ps1 `
  -Configuration Release `
  -WithPaddleOcr `
  -PaddleInferenceDir .\src\BaoBaoPaddleOCR.NET\paddle_inference

# 4. 跑 mock 自检
.\src\BaoBaoPaddleOCR.NET\smoke-test.ps1 -ImagePath .\src\BaoBaoPaddleOCR.NET\PaddleOCR-3.4.0\test_tipc\web\test.jpg