how to run the inference in pure C code?

[ardeal]

I have already trained yolov8 in python code, and have already converted the pth model to onnx model.
I would like to run the inference of the onnx mode in pure C code, how to run it? is there any example code, workflow or document?

[pranavsharma]

https://github.com/microsoft/onnxruntime-inference-examples/tree/main/c_cxx/fns_candy_style_transfer

[ardeal]

Thank you @pranavsharma !

in your examples, you are using exe to run inference.
what I would like to do are:

1. pure C code inference.
2. run on visual studio and embedded MCU respectively.
   so, I would like to check the pure c code and examples about how to run it.

[pranavsharma]

The pure C code is in the same folder in file https://github.com/microsoft/onnxruntime-inference-examples/blob/main/c_cxx/fns_candy_style_transfer/fns_candy_style_transfer.c. You can run cmake using the example CMakeLists.txt file and it'll generate a platform dependent executable that you can run.

[ardeal]

Thank you @pranavsharma ,
The C code in fns_candy_style_transfer.c is not what I am looking for.

I would like to check the operators(i.e. conv2d, resize, linear and etc) C code.

according to the CMakeList.txt file, I guess the C code is in the library onnxruntime. However, I could not fiind the operators C code in onnxruntime after I checked onnxruntime in repo: https://github.com/microsoft/onnxruntime.

I would like to copy the operators C code to embedded MCU and call them with similar C code like fns_candy_style_transfer.c

[pranavsharma]

Code for all the operators are here https://github.com/microsoft/onnxruntime/tree/main/onnxruntime/core/providers/cpu.

[ardeal]

Than you @pranavsharma !

The code in https://github.com/microsoft/onnxruntime/tree/main/onnxruntime/core/providers/cpu is C++ code, but not C code.
What I am looking for is C code.

[pranavsharma]

Sorry, this is a C++ project. We don't have C code.

[ardeal]

in onnxruntime-inference-examples, fns_candy_style_transfer.exe <model_path> <input_image_path> <output_image_path> [cpu|cuda|dml] is a C file, right?

The upper exe will call onnxruntime lib which is a C++.
This confused me. Could the C exe call the C++ lib?

[pranavsharma]

An executable links with the ORT .so/DLL (implicitly [at link time] or explicitly [by calling dlopen]) and uses the C APIs exported by it. Do you have a concrete use case?

[ardeal]

Thank you @pranavsharma !

I really cannot understand. your operators are C++ code, while the exported DLL/APIs are C. it means that you are using C to call C++ functions in the dll. is my understanding correct?

[pranavsharma]

Please familiarize yourself with how shared libs work. This is not an ORT discussion/issue.