C++ Run torch_script model on cv::Mat

[elduderino15 (ElDuderino)]

How do I run a python trained, dumped torch_script model on a cv::Mat of the right input size in C++?
My tries (doing this for the first time) fail with weirdest errors…
and I seem not to find a proper solution. Thank you, EL

[ptrblck]

This topic should have some valid approaches to convert a cv::Mat to a tensor.
You could try to resize the cv::Mat to the right spatial size in OpenCV and add a batch dimension in libtorch (or create a batch of image tensors).

[elduderino15 (ElDuderino)]

Thank you. I am asking because my build system / linker fails (below). Not sure what the proper way is.

C10 error when converting cv::Mat to tensor with from_blob

When I add from_blob to my code the project fails to compile. I want to create a tensor out of a cv::Mat but it keeps failing with the error below. Again, without from_blob my project compiles fine. cv::Mat amplitudePatch = amplitude(roi); torch::from_blob(amplitudePatch.data, {1, amplitudePatch.size().height, amplitudePatch.size().width, amplitudePatch.channels()}, at::TensorOptions(at::kFloat)); SunlightNoiseClassifier/libSunlightNoiseClassifier.a(SunlightNoiseClassifier.cpp.o): In function…