Visualizing Activation Layers

[sidgan]

Visualizations of the Activation Layers should happen after a convolution layer, and not the activation layer itself.

In https://github.com/sicara/tf-explain/blob/master/examples/core/activations_visualization.py#L20

 grid = explainer.explain(data, model, ['activation_6'])

needs to be replaced by

 grid = explainer.explain(data, model, ['conv_1']) 

Happy to send a pull request for the same.

[RaphaelMeudec]

Indeed, but this could have no effect if the activation is defined inside the Conv2D layer. If you run the below snippet, first conv output is the output of the activation layer. Second one is the output of the convolution itself.

Maybe a more appropriate line would be grid = explainer.explain(data, model, ['target_layer'])? What do you think?

Thanks for your feedback by the way!

[sidgan]

Agree with your suggestion.