Unguided Grad CAM

[RaphaelMeudec]

Atm, guided backpropagation is included in Grad CAM automatically (see here). The objective of this issue is to make optional this behaviour (but selected by default).

How to do so?

• If you're not at ease with what Guided Backpropagation is, check the Class Activation Maps section in this article
• Add a use_guided argument (boolean, default value True) to the .explain() method of Grad CAM
• Pass the argument to the get_gradients_and_filters() method
• Compute guidance only is use_guided is True
• Add the argument use_guided to the corresponding callback, and pass it as argument to the explain() method
• Add unit tests

When opening your PR, make sure to add some resulting visualizations.

[RiyazAina-DeepML]

NameError: name 'GradCAM' is not defined

[RaphaelMeudec]

@RiyazAina-DeepML Could you create a new issue and put some context in it? For example, a sample code to reproduce the error. Thanks