I wanted an ONNX runtime example that uses pure C, but without all the slop in the official pure C example. Additionally, I wanted something that works with mingw, so I don't use onnxruntime.lib. Basically, this depends on onnxruntime.dll and the associated headers.
This example just runs a tiny network dumped from PyTorch.
First, ensure that gcc is available on your PATH.
Next, run buid_windows.bat from this directory.
Finally, run onnxruntime_example.exe from this directory. (It will attempt to
load onnxruntime.dll from the onnxruntime\lib subdirectory.)
The example neural network was generated using the generate_network.py
script. To generate a new network, simply run the script and wait for it to
complete. However, you'll need to change the hardcoded test_input and
expected_output portions in onnxruntime_example.c to match the values
printed at the end of the script.
The network simply takes a vector of 4 values and returns a vector approximating two values: the sum of all 4 inputs, and the maximum difference between any two of the inputs.