Using GradCAM for a decoder

[jcjlin]

Hi, I'm trying to use GradCAM in an autoencoder to visualize the feature map of each layers from the encoder and decoder, I successfully got the image from encoder, but for decoder because the input is a tensor but not image, is it avalible to use GradCAM to visuallize a decoder's feature map?

[jacobgil]

Yes it should work !
As long as the tensor has spatial dimensions (width and height).
What are the dimensions of the input tensor ?