Releases: sdcb/PaddleSharp
3.0.1 - Linux/macOS supported
We are excited to announce the release of PaddleSharp v3.0.1!
This version brings official support for Linux and macOS runtime dynamic libraries, further expanding multi-platform capability and bringing you a more robust Cross-Platform Paddle Inference experience.
🆕 Newly Released NuGet Runtime Packages for Linux/macOS
In addition to our existing Windows runtime packages, v3.0.1 introduces six new local dynamic library NuGet packages, enabling support for both x64 and ARM64 architectures on Linux and macOS:
With these additions, PaddleSharp now covers mainstream deployment scenarios for .NET on Windows, Linux, and macOS with both x64 and arm64 support.
📦 Core Package Updates
-
Sdcb.PaddleInference 3.0.1
Maintenance update ensuring compatibility with new runtime libraries. -
Sdcb.PaddleOCR 3.0.1 (Recommended for upgrade!)
- Fixed precision issues and rare crash bugs in OCR models.
⚠️ Note: If you've never installed theSdcb.PaddleOCR
NuGet package directly,
please install or upgrade to this package. (If you previously only usedSdcb.PaddleOCR.Models.Online
orSdcb.PaddleOCR.Models.Local
, your OCR backend was still using an older version.)
-
Sdcb.RotationDetector 1.0.4
- Updated dependency to
Sdcb.PaddleInference 3.0.1
(no other changes).
- Updated dependency to
-
Sdcb.PaddleNLP.Lac 1.0.1
- Updated dependency to
Sdcb.PaddleInference 3.0.1
(no other changes).
- Updated dependency to
💡 Package Selection Guide
The complete set of runtime packages—including Windows, Linux, and macOS libraries—are self-built with open-source enhancements. You can find all source code and CI scripts in the PaddleSharp GitHub repository.
🙌 Thanks for Your Support!
Thank you for your continued feedback and support!
As always, we welcome your suggestions and pull requests, and we hope you enjoy PaddleSharp v3.0.1 on all major platforms!
3.0
PaddleSharp v3.0 Official Release
We are excited to announce the official release of PaddleSharp version 3.0! This new version supports the Paddle Inference Library 3.0 as well as the latest PaddleOCR v5 models. We welcome everyone to try it out!
Newly released .NET NuGet packages (5):
Sdcb.PaddleInference.3.0.0
- Supports Paddle Inference Library C API version 3.0Sdcb.PaddleOCR.3.0.0
- Supports PaddleOCR v5 modelsSdcb.PaddleOCR.Models.Local.3.0.1
Sdcb.PaddleOCR.Models.LocalV5.3.0.0
- Includes local models of PaddleOCR v5Sdcb.PaddleOCR.Models.Online.3.0.1
- Provides auto-downloading for PaddleOCR v5 models
In addition, 16 runtime-specific NuGet packages for Paddle Inference have been released:
Sdcb.PaddleInference.runtime.win64.mkl
Sdcb.PaddleInference.runtime.win64.openblas
Sdcb.PaddleInference.runtime.win64.openblas-noavx
Sdcb.PaddleInference.runtime.win64.cu118_cudnn89_sm61
Sdcb.PaddleInference.runtime.win64.cu118_cudnn89_sm75
Sdcb.PaddleInference.runtime.win64.cu118_cudnn89_sm86
Sdcb.PaddleInference.runtime.win64.cu118_cudnn89_sm89
Sdcb.PaddleInference.runtime.win64.cu126_cudnn95_sm61
Sdcb.PaddleInference.runtime.win64.cu126_cudnn95_sm75
Sdcb.PaddleInference.runtime.win64.cu126_cudnn95_sm86
Sdcb.PaddleInference.runtime.win64.cu126_cudnn95_sm89
Sdcb.PaddleInference.runtime.win64.cu129_cudnn910_sm61
Sdcb.PaddleInference.runtime.win64.cu129_cudnn910_sm75
Sdcb.PaddleInference.runtime.win64.cu129_cudnn910_sm86
Sdcb.PaddleInference.runtime.win64.cu129_cudnn910_sm89
Sdcb.PaddleInference.runtime.win64.cu129_cudnn910_sm120
Package Selection Guide:
- We recommend
Sdcb.PaddleInference.runtime.win64.mkl
for most users. It offers the best balance between performance and package size. Please note that this package does not support GPU acceleration, making it suitable for most general scenarios. openblas-noavx
is tailored for older CPUs that do not support the AVX2 instruction set.- The remaining packages cover various CUDA combinations (GPU acceleration), supporting three CUDA versions:
- CUDA 11.8: Supports 10–40 series NVIDIA GPUs
- CUDA 12.6: Supports 10–40 series NVIDIA GPUs
- CUDA 12.9: Supports 10–50 series NVIDIA GPUs
Important:
Not all GPU packages are suitable for every card. Please refer to the following GPU-to-sm
suffix mapping:
sm Suffix |
Supported GPU Series |
---|---|
sm61 | GTX 10 Series |
sm75 | RTX 20 Series (and GTX 16xx series such as GTX 1660) |
sm86 | RTX 30 Series |
sm89 | RTX 40 Series |
sm120 | RTX 50 Series (supported by CUDA 12.9 only) |
All NuGet packages are self-compiled, with minor modifications to the official C API libraries. Both the changes and the build process are fully open-source. You are welcome to star the PaddleSharp open-source project and provide your feedback or suggestions!
2.5.0.1
What's Changed
$ git ls-files -z | xargs -0 sed -i 's/Size.Zero/default/'
by @n0099 in #78- Feature/2.6, added PaddleNLP.Lac by @sdcb in #86
- Chore: Add issue template files by @JeremyWu917 in #89
- Update PaddleDetector.cs by @LuGuangguang in #91
- PaddleOCR 2.7.0.2: Supports thread-safe.
New Contributors
- @JeremyWu917 made their first contribution in #89
- @LuGuangguang made their first contribution in #91
Full Changelog: 2.5.0...2.5.0.1
2.5.0
2.4.1.5
Latest before merge 2.5
Full Changelog: 2.4.1.4...2.4.1.5
2.4.1.4
2.4.1.3
Published following packages:
Release notes:
- All packages add detailed xml comments to all public properties, methods, classes, enums.
- All packages included xml comments when packaging
- All packages added .snuget symbol files, can dig into source code when debugging.
2.4.1.2
PaddleSharp有个重磅更新发布啦!
✅ Sdcb.PaddleInference 2.4.1.2
✅ Sdcb.PaddleOCR 2.6.0.4
✅ Sdcb.PaddleOCR.Models.LocalV3 2.6.0.3
✅ Sdcb.PaddleOCR.Models.Online 2.6.0.3
这次更新增加了基于百度深度学习🧠的表格检测模型,进行更强大的表格识别功能!表格支持跨行和跨列哦📊📈!📚使用示例完整代码及文档,点击这里获取:https://github.com/sdcb/PaddleSharp/blob/master/docs/ocr.md#table-recognition
💥除此之外,还发布了基于Ubuntu 22.04的新docker镜像,让Linux下使用更加方便:
🔹 sdflysha/dotnet6-paddle:2.4.2-ubuntu22
🔹 sdflysha/dotnet6sdk-paddle:2.4.2-ubuntu22
Github仓库地址(欢迎给我们点个star哦):https://github.com/sdcb/PaddleSharp
📦 NuGet包下载:https://www.nuget.org/packages?q=Sdcb.Paddle
一些细节方面的优化:
- 将Mkldnn默认大小从10改到了1(感谢n0099大佬)
- 将OCR结果进行了“合理化”的排序,以后OCR结果将不再会颠倒过来了
- 将QueuedPaddleOcrAll默认队列数量从4改为了1,更适合预期
- 同时将表格识别模型加入了本地NuGet包和在线下载NuGet包,本地NuGet包大小从106MB提高到119MB
- 将OpenCvSharp4依赖升级到了最新
Full Changelog: 2.4.1.1...2.4.1.2
2.4.1.1
Versions(版本号)
Sdcb.PaddleOCR 2.6.0.2
Sdcb.PaddleOCR.Models.LocalV3 2.6.0
Sdcb.PaddleOCR.Models.Online 2.6.0
Sdcb.PaddleInference 2.4.1.1
Addressed following issues/added following features(解决了如下问题)
- PaddleOCR可以走多线程队列了 PaddleOCR can run as multi-threaded queue now!
code examples(代码示例): https://github.com/sdcb/PaddleSharp/blob/9e95f07417a513b0e7d0cc1fb3ff898d6e4a346e/tests/Sdcb.PaddleOCR.Tests/OfflineModelsTest.cs#L51
implementing details(实现细节): https://github.com/sdcb/PaddleSharp/blob/master/src/Sdcb.PaddleOCR/QueuedPaddleOcrAll.cs
Running as a queued can avoid crash when using multiple thread GPU inference, can also maximizing GPU performance to the extreme
(使用走队列的GPU推理可以避免Web项目走多线程GPU时爆炸的问题,也可以将GPU的潜力压榨到极致) - 为LocalV3和Online模型添加了更多语言的支持,Add more language model combination, thanks n0099: #35
- useMultiStream=false by default when use Gpu(使用GPU时,将useMultiStream改为了false)
- Improved annoying console warning message when UseMultiStream(改善了UseMultiStream时的控制台警告提示)
What's Changed
- add more combinations of
det+cls+rec
models for all available languages @Sdcb.PaddleOCR.Models.Online.OnlineFullModels
by @n0099 in #35
New Contributors
Full Changelog: 2.4.0...2.4.1.1