Efficientnet classifier interpretation using GradCAM - part 3 of 3

less than 1 minute read

In the previous article, we managed to build a binary classifier to detect if patient had a pneumothorax or not. Neural networks commonly suffer from a lack of explainability. In this project, we will use a method called Grad-CAM (Gradient-weighted Class Activation Mapping) to visualize heatmaps on xray corresponding to the most important zones of the image for our classifier.

What is Grad-CAM ?