Hi,
I built a small streamlit app that uses various methods of Class Activation Maps to show what parts of an image contributes to a models (ResNet18 trained on ImageNet) predictions. You can also choose a category to see what parts of the image contribute to the model predicting that category specifically.