search

PAIR-code / saliency

Framework-agnostic implementation for state-of-the-art saliency methods (XRAI, BlurIG, SmoothGrad, and more).
https://pair-code.github.io/saliency/
Apache License 2.0
950 stars 191 forks source link

New notebook Examples_pytorch.ipynb, using torchvision pretrained model #63

Closed bwedin closed 3 years ago

bwedin commented 3 years ago

Adds new notebook to base directory "Examples_pytorch.ipynb", which calls an InceptionV3 PyTorch model (from pretrained models in torchvision package)

bwedin commented 3 years ago

I am not an expert in PyTorch but would it be better to use torch.movedim(...) on the outputs of interest directly instead of registering hooks?

What do you mean? It looks like this function could help with having to change between [3,W,H] and [W,H,3] shapes, but I don't see how that stops us from having to use the hooks?