Explainable AI using Saliency maps and Proto-PNet

class activation mappings and prototype-based networks for explanations in CNN classifications

Class Activation mappings (CAM) based on the output of a simple ResNet18 model

Prototype based network. Works on a this-looks-like-that principle, i.e. it looks in other images with similar features to explain why it classifies image to certain classes. Proto-PNet model architecture

Explanations from the Proto-PNet model