ProtoPNet: save learned prototype patches

[gpitteri]

So far the learned prototype patches are just visualised at the end of the training and not stored (otherwise we will end up with 2000 saved png images). We need to decide how we want to handle this.

Furthermore the visualization is done only if we run the prototype explanation first during the evaluation of the model.

[gpitteri]

The same problem during the evaluation of the model. For each test image we visualise the top k activated prototypes. This only works if we evaluated the prototypes and there is no saving feature of the prototypes.