4bb9bc738c
* change docs from 1.x to main * bump version * update readme status info and highlights * revert version change * fix dead links * fix script * fix docs * fix building status * fix md lint error * upgrade to ubuntu20.04 for github ci * Revert "upgrade to ubuntu20.04 for github ci" This reverts commit 9c598f4857633dfb96b116ca9fb9a5d15b5fa6b4. * update codebase md files * upgrade to ubuntu20.04 for github ci * fix * fix requirements * fix * fix ci * bump version * fix * fix pplnn torchscript * fix building status * fix ncnn * fix * remove circileci hold * fix pytorch image
Usage
step 0. install the local nuget package
You should build csharp api first, it will generate a nuget package, or you can download our prebuit package. You may refer to this on how to install local nuget package.
step 1. Add runtime dll to the system path
If you built csharp api from source and didn't build static lib, you should add the built dll to your system path. The same is to opencv, etc.
And don't forget to install backend dependencies. Take tensorrt backend as example, you have to install cudatoolkit, cudnn and tensorrt. The version of backend dependencies that our prebuit nuget package used will be offered in release note.
| backend | dependencies |
|---|---|
| tensorrt | cudatoolkit, cudnn, tensorrt |
| onnxruntime | onnxruntime / onnxruntime-gpu |
step 2. Open Demo.sln and build solution.
step 3. Prepare the model.
You can either convert your model according to this tutorial or download the test models from OneDrive or BaiduYun. The web drive contains onnx and tensorrt models and the test models are converted under environment of cuda11.1 + cudnn8.2.1 + tensorrt 8.2.3.0 + GTX2070s.
Note:
- a) If you want to use the tensorrt model from the link, make sure your environment and your gpu architecture is same with above.
- b) When you use the downloaded onnx model, you have to edit
deploy.json, editend2end.enginetoend2end.onnxandtensorrttoonnxruntime.
step 4. Set one project as startup project and run it.