[CppDay20] Interoperable AI: ONNX & ONNXRuntime in C++ (M. Arena, M.Verasani)

[@Chendadon]

This is gold! Thank you very much! I'll be happy to see more content like this. And if you have more resources and examples on ONNX please share

[@ilpropheta]

Thanks for your kind words!

[@Daniel152315]

A really helpful video, even now in 2022! Thank you! They still need to work on the C++ docs and create a couple tutorials, it is tough to get started but you really helped me

[@ilpropheta]

Very glad to hear that! Thanks for sharing

[@Daniel152315]

@@ilpropheta Hey, you mentioned wrappers for ONNX Runtime in the video. Could you recommend any easy to use C++ wrappers? Thanks

[@ilpropheta]

@@Daniel152315 Hi Daniel, my apologies, I had replied 2 weeks ago but I don't know why the comment has not been published...I don't know any C++ OnnxRuntime wrapper except for the official library. At my company, I did one myself (cannot release it publicly, I have asked though). My recommendation is to wrap the functionalities you use the most and refactor all the boilerplate code into some common units. For example, I have a generic wrapper for the session that takes data as std::span and makes all the tensors automatically, returning the results.

[@sinangok2656]

Thank you for the presentation. It was very informative. 👏

[@ilpropheta]

Thanks for watching and for your kind words!

[@galdavid6878]

Thanks for the helpful video! I am interested in adding a new custom operator to onnxruntime in C++, but haven't found a useful tutorial for it yet. Do you know maybe an example/tutorial that can help me with that?

[@ilpropheta]

Thanks! It's a very good question. Mattia and I never added custom operators but this is something we will need soon. I think I would start from the official documentation on GitHub since I am not aware of other tutorials on this topic. If anyone else here knows one, please let us know!