Closed arasharchor closed 6 years ago
Hi @smajida Yes, it is most certainly possible. We had that in mind when designing tf_cnnvis, so we do have provision for fully connected network too. This will show you a kind of heatmap of the reconstructed input- indicating which nodes are more important. This is the closest thing we could think of, as reconstructing "image" per se is not an option for 1-d inputs. We have tested a simple case for activation visualization. If you use it in a prototype model which can be shared, please share or add a notebook in the examples. Thanks!
@falaktheoptimist thanks for your response I am working on it with my colleague. I will update you as soon as I have something to show.
Hi,
i have mostly seen people use visualization for image recognition models. Does it make sense and is it possible to use your code for visualizing FCN network layers?