Pytorch c++ Grad-CAM

Does someone already try to do Grad-CAM (heatmap) with pytorch c++?

For now, I did the training in python and load the model with:

m_module = torch::jit::load(pathscript);

I’m pretty sure that I can’t use jit trace to do Grad-CAM and I will need to redefine my CNN in c++ directly.

struct Net: torch::nn::Module

Can someone confirm?


1 Like

I am very curious about this as well. Any luck?