I want to perform transfer learning using Libtorch (C++).
At that time, I think about using a trained model (for example, 1000 classifications of Resnet50), but I want to change the number of output classifications each time.
Please tell me how to change the dimension of the final layer of torchscript only with Libtorch for testing in a C++ environment without Python.
As long as you can save the dimension-changed model with jit::model, I can handle the rest.
that’s all. thank you.