Extract feature representation of each layer during training

Hi, I am working on the implementation of a certain loss that requires the extraction of the feature maps of each layer in a Fully ConvNet to calculate the new loss, it is possible to do that? I know that is possible to extract the feature map after the training, but, it’s possible to do it during training without stopping it?

It can be done easily with the forward hooks. Check this answer of mine for more details.

1 Like

This is what i was looking for, thank you!