Computing Saliency Map or Integrated Gradients with pretrained model

vision
1

I have pretrained network, let’s say patchGAN and I want to compute the saliency map of the images passed through the pretrained network. How can I achieve it.

I tried different scripts but it didn’t help.