Different behaivour of exported model in TorchScript

jit
1

Hi there,

I am trying to export and use a model defined in pytorch by using the “torch.jit.script” method. When I try to use the exported model the result obtained is awful if you compare with the original results. The model loads pretrained weigths and after all of these I use the method “torch.jit.script” and save it to the file. I have found some related issue in github, however I have not managed to solve the problem.

Any idea of the problem? Thank you in advance.

2

Do you have some code to reproduce the issue? It’s hard to say what’s going on without more information

3

Thank you for your response. Finally, I have managed to get similar results using the exported model. The problem was that the tensor was transposed by using a custom ToTensor transform instead of using the standard ToTensor transform.