Benchmark other onnx runtimes

[diegobernardes]

Is your feature request related to a problem? Please describe.
Would be nice to know the runtime performance of onnx-go compared with the other runtimes available. Here a list of some runtimes: https://onnx.ai/supported-tools

Describe the solution you'd like
Don't need to be nothing fancy, maybe a wiki page with some models being compared between the other implementations.

For some solutions, mainly edge, performance is really a key factor, this could be a really advantage to onnx-go because of the Go speed and concurrency. It will also highlight the parts we need to dedicate more effort to make it faster.

[owulveryck]

I fully agree.
The key point is to compare against Microsoft's onnxruntime.
The goal of onnxruntime is similar: to run pre-trained neural network without pain.

Other comparisons are also interesting. For info, there is a trivial bench in the examples/model_zoo_executor that runs a model against whatever onnx model referenced by the $MODELDIR environment variable.

[owulveryck]

BTW, I had a discussion with @xadupre a while ago.
He mentioned one interesting project: https://github.com/sdpython/scikit-learn_benchmarks which is using asv (I love the name of the project...)

I did not have a closer look at the project yet, but if the effort is low, it could be a good idea to use asv for the benchmark.

What do you think?