ofxOrt

Thin wrapper of ONNX runtime for openFrameworks.

Note

This addon is still in an experimental stage, and I'm not skilled developer, so I'll make breaking changes frequently :(

Dependencies

• ofxCv
• ofxOpenCV

Prepare

Install ONNX runtime by following the document.

Environment

Author's environment is

• Windows10
• NVIDIA Driver:462.80
• CUDA 11.0.221
• cuDNN: 8.0.2.39
• Visual studio 2019
• oF:11.0

I think it may also work in other environments, such as Linux and Mac, but I have not tested.(I'll be verifying this soon.)

Usage

Prepare ONNX model and copy it to /bin

Windows

If you use CUDA on windows, please copy onnxruntime_providers_cuda.dll and onnxruntime_providers_shared.dll to /bin.

//First, create ofxOrt instance
ofxOrt ofxOrt(ORT_TSTR("model.name");

//Create input and output tensors
auto memory_info = Ort::MemoryInfo::CreateCpu(OrtDeviceAllocator, OrtMemTypeCPU);
ofxOrtImageTensor<float> input_tensor(memory_info, content.getTexture());
ofxOrtImageTensor<float> output_tensor(memory_info, numChannels, outWidth, outHeight, true);

//Run inference
ort.forward(Ort::RunOptions{ nullptr }, input_names, &(input_tensor.getTensor()), 1, output_names, &(output_tensor.getTensor()), 1);

References

ONNX Runtime Inference Examples onnx_runtime_cpp

License

ofxOrt is under MIT license.